src/cpu/sparc/vm/templateTable_sparc.cpp

Fri, 29 Apr 2011 11:15:30 -0700

author
kvn
date
Fri, 29 Apr 2011 11:15:30 -0700
changeset 2816
286c498ae0d4
parent 2811
08ccee2c4dbf
parent 2784
92add02409c9
child 2982
ddd894528dbc
permissions
-rw-r--r--

Merge

     1 /*
     2  * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12  * version 2 for more details (a copy is included in the LICENSE file that
    13  * accompanied this code).
    14  *
    15  * You should have received a copy of the GNU General Public License version
    16  * 2 along with this work; if not, write to the Free Software Foundation,
    17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18  *
    19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    20  * or visit www.oracle.com if you need additional information or have any
    21  * questions.
    22  *
    23  */
    25 #include "precompiled.hpp"
    26 #include "interpreter/interpreter.hpp"
    27 #include "interpreter/interpreterRuntime.hpp"
    28 #include "interpreter/templateTable.hpp"
    29 #include "memory/universe.inline.hpp"
    30 #include "oops/methodDataOop.hpp"
    31 #include "oops/objArrayKlass.hpp"
    32 #include "oops/oop.inline.hpp"
    33 #include "prims/methodHandles.hpp"
    34 #include "runtime/sharedRuntime.hpp"
    35 #include "runtime/stubRoutines.hpp"
    36 #include "runtime/synchronizer.hpp"
    38 #ifndef CC_INTERP
    39 #define __ _masm->
    41 // Misc helpers
    43 // Do an oop store like *(base + index + offset) = val
    44 // index can be noreg,
    45 static void do_oop_store(InterpreterMacroAssembler* _masm,
    46                          Register base,
    47                          Register index,
    48                          int offset,
    49                          Register val,
    50                          Register tmp,
    51                          BarrierSet::Name barrier,
    52                          bool precise) {
    53   assert(tmp != val && tmp != base && tmp != index, "register collision");
    54   assert(index == noreg || offset == 0, "only one offset");
    55   switch (barrier) {
    56 #ifndef SERIALGC
    57     case BarrierSet::G1SATBCT:
    58     case BarrierSet::G1SATBCTLogging:
    59       {
    60         // Load and record the previous value.
    61         __ g1_write_barrier_pre(base, index, offset,
    62                                 noreg /* pre_val */,
    63                                 tmp, true /*preserve_o_regs*/);
    65         if (index == noreg ) {
    66           assert(Assembler::is_simm13(offset), "fix this code");
    67           __ store_heap_oop(val, base, offset);
    68         } else {
    69           __ store_heap_oop(val, base, index);
    70         }
    72         // No need for post barrier if storing NULL
    73         if (val != G0) {
    74           if (precise) {
    75             if (index == noreg) {
    76               __ add(base, offset, base);
    77             } else {
    78               __ add(base, index, base);
    79             }
    80           }
    81           __ g1_write_barrier_post(base, val, tmp);
    82         }
    83       }
    84       break;
    85 #endif // SERIALGC
    86     case BarrierSet::CardTableModRef:
    87     case BarrierSet::CardTableExtension:
    88       {
    89         if (index == noreg ) {
    90           assert(Assembler::is_simm13(offset), "fix this code");
    91           __ store_heap_oop(val, base, offset);
    92         } else {
    93           __ store_heap_oop(val, base, index);
    94         }
    95         // No need for post barrier if storing NULL
    96         if (val != G0) {
    97           if (precise) {
    98             if (index == noreg) {
    99               __ add(base, offset, base);
   100             } else {
   101               __ add(base, index, base);
   102             }
   103           }
   104           __ card_write_barrier_post(base, val, tmp);
   105         }
   106       }
   107       break;
   108     case BarrierSet::ModRef:
   109     case BarrierSet::Other:
   110       ShouldNotReachHere();
   111       break;
   112     default      :
   113       ShouldNotReachHere();
   115   }
   116 }
   119 //----------------------------------------------------------------------------------------------------
   120 // Platform-dependent initialization
   122 void TemplateTable::pd_initialize() {
   123   // (none)
   124 }
   127 //----------------------------------------------------------------------------------------------------
   128 // Condition conversion
   129 Assembler::Condition ccNot(TemplateTable::Condition cc) {
   130   switch (cc) {
   131     case TemplateTable::equal        : return Assembler::notEqual;
   132     case TemplateTable::not_equal    : return Assembler::equal;
   133     case TemplateTable::less         : return Assembler::greaterEqual;
   134     case TemplateTable::less_equal   : return Assembler::greater;
   135     case TemplateTable::greater      : return Assembler::lessEqual;
   136     case TemplateTable::greater_equal: return Assembler::less;
   137   }
   138   ShouldNotReachHere();
   139   return Assembler::zero;
   140 }
   142 //----------------------------------------------------------------------------------------------------
   143 // Miscelaneous helper routines
   146 Address TemplateTable::at_bcp(int offset) {
   147   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
   148   return Address(Lbcp, offset);
   149 }
   152 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register Rbyte_code,
   153                                    Register Rscratch,
   154                                    bool load_bc_into_scratch /*=true*/) {
   155   // With sharing on, may need to test methodOop flag.
   156   if (!RewriteBytecodes) return;
   157   if (load_bc_into_scratch) __ set(bc, Rbyte_code);
   158   Label patch_done;
   159   if (JvmtiExport::can_post_breakpoint()) {
   160     Label fast_patch;
   161     __ ldub(at_bcp(0), Rscratch);
   162     __ cmp(Rscratch, Bytecodes::_breakpoint);
   163     __ br(Assembler::notEqual, false, Assembler::pt, fast_patch);
   164     __ delayed()->nop();  // don't bother to hoist the stb here
   165     // perform the quickening, slowly, in the bowels of the breakpoint table
   166     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, Rbyte_code);
   167     __ ba(false, patch_done);
   168     __ delayed()->nop();
   169     __ bind(fast_patch);
   170   }
   171 #ifdef ASSERT
   172   Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
   173   Label okay;
   174   __ ldub(at_bcp(0), Rscratch);
   175   __ cmp(Rscratch, orig_bytecode);
   176   __ br(Assembler::equal, false, Assembler::pt, okay);
   177   __ delayed() ->cmp(Rscratch, Rbyte_code);
   178   __ br(Assembler::equal, false, Assembler::pt, okay);
   179   __ delayed()->nop();
   180   __ stop("Rewriting wrong bytecode location");
   181   __ bind(okay);
   182 #endif
   183   __ stb(Rbyte_code, at_bcp(0));
   184   __ bind(patch_done);
   185 }
   187 //----------------------------------------------------------------------------------------------------
   188 // Individual instructions
   190 void TemplateTable::nop() {
   191   transition(vtos, vtos);
   192   // nothing to do
   193 }
   195 void TemplateTable::shouldnotreachhere() {
   196   transition(vtos, vtos);
   197   __ stop("shouldnotreachhere bytecode");
   198 }
   200 void TemplateTable::aconst_null() {
   201   transition(vtos, atos);
   202   __ clr(Otos_i);
   203 }
   206 void TemplateTable::iconst(int value) {
   207   transition(vtos, itos);
   208   __ set(value, Otos_i);
   209 }
   212 void TemplateTable::lconst(int value) {
   213   transition(vtos, ltos);
   214   assert(value >= 0, "check this code");
   215 #ifdef _LP64
   216   __ set(value, Otos_l);
   217 #else
   218   __ set(value, Otos_l2);
   219   __ clr( Otos_l1);
   220 #endif
   221 }
   224 void TemplateTable::fconst(int value) {
   225   transition(vtos, ftos);
   226   static float zero = 0.0, one = 1.0, two = 2.0;
   227   float* p;
   228   switch( value ) {
   229    default: ShouldNotReachHere();
   230    case 0:  p = &zero;  break;
   231    case 1:  p = &one;   break;
   232    case 2:  p = &two;   break;
   233   }
   234   AddressLiteral a(p);
   235   __ sethi(a, G3_scratch);
   236   __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
   237 }
   240 void TemplateTable::dconst(int value) {
   241   transition(vtos, dtos);
   242   static double zero = 0.0, one = 1.0;
   243   double* p;
   244   switch( value ) {
   245    default: ShouldNotReachHere();
   246    case 0:  p = &zero;  break;
   247    case 1:  p = &one;   break;
   248   }
   249   AddressLiteral a(p);
   250   __ sethi(a, G3_scratch);
   251   __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
   252 }
   255 // %%%%% Should factore most snippet templates across platforms
   257 void TemplateTable::bipush() {
   258   transition(vtos, itos);
   259   __ ldsb( at_bcp(1), Otos_i );
   260 }
   262 void TemplateTable::sipush() {
   263   transition(vtos, itos);
   264   __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
   265 }
   267 void TemplateTable::ldc(bool wide) {
   268   transition(vtos, vtos);
   269   Label call_ldc, notInt, notString, notClass, exit;
   271   if (wide) {
   272     __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
   273   } else {
   274     __ ldub(Lbcp, 1, O1);
   275   }
   276   __ get_cpool_and_tags(O0, O2);
   278   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
   279   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
   281   // get type from tags
   282   __ add(O2, tags_offset, O2);
   283   __ ldub(O2, O1, O2);
   284   __ cmp(O2, JVM_CONSTANT_UnresolvedString);    // unresolved string? If so, must resolve
   285   __ brx(Assembler::equal, true, Assembler::pt, call_ldc);
   286   __ delayed()->nop();
   288   __ cmp(O2, JVM_CONSTANT_UnresolvedClass);     // unresolved class? If so, must resolve
   289   __ brx(Assembler::equal, true, Assembler::pt, call_ldc);
   290   __ delayed()->nop();
   292   __ cmp(O2, JVM_CONSTANT_UnresolvedClassInError);     // unresolved class in error state
   293   __ brx(Assembler::equal, true, Assembler::pn, call_ldc);
   294   __ delayed()->nop();
   296   __ cmp(O2, JVM_CONSTANT_Class);      // need to call vm to get java mirror of the class
   297   __ brx(Assembler::notEqual, true, Assembler::pt, notClass);
   298   __ delayed()->add(O0, base_offset, O0);
   300   __ bind(call_ldc);
   301   __ set(wide, O1);
   302   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), O1);
   303   __ push(atos);
   304   __ ba(false, exit);
   305   __ delayed()->nop();
   307   __ bind(notClass);
   308  // __ add(O0, base_offset, O0);
   309   __ sll(O1, LogBytesPerWord, O1);
   310   __ cmp(O2, JVM_CONSTANT_Integer);
   311   __ brx(Assembler::notEqual, true, Assembler::pt, notInt);
   312   __ delayed()->cmp(O2, JVM_CONSTANT_String);
   313   __ ld(O0, O1, Otos_i);
   314   __ push(itos);
   315   __ ba(false, exit);
   316   __ delayed()->nop();
   318   __ bind(notInt);
   319  // __ cmp(O2, JVM_CONSTANT_String);
   320   __ brx(Assembler::notEqual, true, Assembler::pt, notString);
   321   __ delayed()->ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
   322   __ ld_ptr(O0, O1, Otos_i);
   323   __ verify_oop(Otos_i);
   324   __ push(atos);
   325   __ ba(false, exit);
   326   __ delayed()->nop();
   328   __ bind(notString);
   329  // __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
   330   __ push(ftos);
   332   __ bind(exit);
   333 }
   335 // Fast path for caching oop constants.
   336 // %%% We should use this to handle Class and String constants also.
   337 // %%% It will simplify the ldc/primitive path considerably.
   338 void TemplateTable::fast_aldc(bool wide) {
   339   transition(vtos, atos);
   341   if (!EnableInvokeDynamic) {
   342     // We should not encounter this bytecode if !EnableInvokeDynamic.
   343     // The verifier will stop it.  However, if we get past the verifier,
   344     // this will stop the thread in a reasonable way, without crashing the JVM.
   345     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
   346                      InterpreterRuntime::throw_IncompatibleClassChangeError));
   347     // the call_VM checks for exception, so we should never return here.
   348     __ should_not_reach_here();
   349     return;
   350   }
   352   Register Rcache = G3_scratch;
   353   Register Rscratch = G4_scratch;
   355   resolve_cache_and_index(f1_oop, Otos_i, Rcache, Rscratch, wide ? sizeof(u2) : sizeof(u1));
   357   __ verify_oop(Otos_i);
   359   Label L_done;
   360   const Register Rcon_klass = G3_scratch;  // same as Rcache
   361   const Register Rarray_klass = G4_scratch;  // same as Rscratch
   362   __ load_klass(Otos_i, Rcon_klass);
   363   AddressLiteral array_klass_addr((address)Universe::systemObjArrayKlassObj_addr());
   364   __ load_contents(array_klass_addr, Rarray_klass);
   365   __ cmp(Rarray_klass, Rcon_klass);
   366   __ brx(Assembler::notEqual, false, Assembler::pt, L_done);
   367   __ delayed()->nop();
   368   __ ld(Address(Otos_i, arrayOopDesc::length_offset_in_bytes()), Rcon_klass);
   369   __ tst(Rcon_klass);
   370   __ brx(Assembler::zero, true, Assembler::pt, L_done);
   371   __ delayed()->clr(Otos_i);    // executed only if branch is taken
   373   // Load the exception from the system-array which wraps it:
   374   __ load_heap_oop(Otos_i, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
   375   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
   377   __ bind(L_done);
   378 }
   380 void TemplateTable::ldc2_w() {
   381   transition(vtos, vtos);
   382   Label retry, resolved, Long, exit;
   384   __ bind(retry);
   385   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
   386   __ get_cpool_and_tags(O0, O2);
   388   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
   389   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
   390   // get type from tags
   391   __ add(O2, tags_offset, O2);
   392   __ ldub(O2, O1, O2);
   394   __ sll(O1, LogBytesPerWord, O1);
   395   __ add(O0, O1, G3_scratch);
   397   __ cmp(O2, JVM_CONSTANT_Double);
   398   __ brx(Assembler::notEqual, false, Assembler::pt, Long);
   399   __ delayed()->nop();
   400   // A double can be placed at word-aligned locations in the constant pool.
   401   // Check out Conversions.java for an example.
   402   // Also constantPoolOopDesc::header_size() is 20, which makes it very difficult
   403   // to double-align double on the constant pool.  SG, 11/7/97
   404 #ifdef _LP64
   405   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
   406 #else
   407   FloatRegister f = Ftos_d;
   408   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset, f);
   409   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset + sizeof(jdouble)/2,
   410          f->successor());
   411 #endif
   412   __ push(dtos);
   413   __ ba(false, exit);
   414   __ delayed()->nop();
   416   __ bind(Long);
   417 #ifdef _LP64
   418   __ ldx(G3_scratch, base_offset, Otos_l);
   419 #else
   420   __ ld(G3_scratch, base_offset, Otos_l);
   421   __ ld(G3_scratch, base_offset + sizeof(jlong)/2, Otos_l->successor());
   422 #endif
   423   __ push(ltos);
   425   __ bind(exit);
   426 }
   429 void TemplateTable::locals_index(Register reg, int offset) {
   430   __ ldub( at_bcp(offset), reg );
   431 }
   434 void TemplateTable::locals_index_wide(Register reg) {
   435   // offset is 2, not 1, because Lbcp points to wide prefix code
   436   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
   437 }
   439 void TemplateTable::iload() {
   440   transition(vtos, itos);
   441   // Rewrite iload,iload  pair into fast_iload2
   442   //         iload,caload pair into fast_icaload
   443   if (RewriteFrequentPairs) {
   444     Label rewrite, done;
   446     // get next byte
   447     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
   449     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
   450     // last two iloads in a pair.  Comparing against fast_iload means that
   451     // the next bytecode is neither an iload or a caload, and therefore
   452     // an iload pair.
   453     __ cmp(G3_scratch, (int)Bytecodes::_iload);
   454     __ br(Assembler::equal, false, Assembler::pn, done);
   455     __ delayed()->nop();
   457     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
   458     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   459     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
   461     __ cmp(G3_scratch, (int)Bytecodes::_caload);
   462     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   463     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
   465     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
   466     // rewrite
   467     // G4_scratch: fast bytecode
   468     __ bind(rewrite);
   469     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
   470     __ bind(done);
   471   }
   473   // Get the local value into tos
   474   locals_index(G3_scratch);
   475   __ access_local_int( G3_scratch, Otos_i );
   476 }
   478 void TemplateTable::fast_iload2() {
   479   transition(vtos, itos);
   480   locals_index(G3_scratch);
   481   __ access_local_int( G3_scratch, Otos_i );
   482   __ push_i();
   483   locals_index(G3_scratch, 3);  // get next bytecode's local index.
   484   __ access_local_int( G3_scratch, Otos_i );
   485 }
   487 void TemplateTable::fast_iload() {
   488   transition(vtos, itos);
   489   locals_index(G3_scratch);
   490   __ access_local_int( G3_scratch, Otos_i );
   491 }
   493 void TemplateTable::lload() {
   494   transition(vtos, ltos);
   495   locals_index(G3_scratch);
   496   __ access_local_long( G3_scratch, Otos_l );
   497 }
   500 void TemplateTable::fload() {
   501   transition(vtos, ftos);
   502   locals_index(G3_scratch);
   503   __ access_local_float( G3_scratch, Ftos_f );
   504 }
   507 void TemplateTable::dload() {
   508   transition(vtos, dtos);
   509   locals_index(G3_scratch);
   510   __ access_local_double( G3_scratch, Ftos_d );
   511 }
   514 void TemplateTable::aload() {
   515   transition(vtos, atos);
   516   locals_index(G3_scratch);
   517   __ access_local_ptr( G3_scratch, Otos_i);
   518 }
   521 void TemplateTable::wide_iload() {
   522   transition(vtos, itos);
   523   locals_index_wide(G3_scratch);
   524   __ access_local_int( G3_scratch, Otos_i );
   525 }
   528 void TemplateTable::wide_lload() {
   529   transition(vtos, ltos);
   530   locals_index_wide(G3_scratch);
   531   __ access_local_long( G3_scratch, Otos_l );
   532 }
   535 void TemplateTable::wide_fload() {
   536   transition(vtos, ftos);
   537   locals_index_wide(G3_scratch);
   538   __ access_local_float( G3_scratch, Ftos_f );
   539 }
   542 void TemplateTable::wide_dload() {
   543   transition(vtos, dtos);
   544   locals_index_wide(G3_scratch);
   545   __ access_local_double( G3_scratch, Ftos_d );
   546 }
   549 void TemplateTable::wide_aload() {
   550   transition(vtos, atos);
   551   locals_index_wide(G3_scratch);
   552   __ access_local_ptr( G3_scratch, Otos_i );
   553   __ verify_oop(Otos_i);
   554 }
   557 void TemplateTable::iaload() {
   558   transition(itos, itos);
   559   // Otos_i: index
   560   // tos: array
   561   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   562   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
   563 }
   566 void TemplateTable::laload() {
   567   transition(itos, ltos);
   568   // Otos_i: index
   569   // O2: array
   570   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   571   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
   572 }
   575 void TemplateTable::faload() {
   576   transition(itos, ftos);
   577   // Otos_i: index
   578   // O2: array
   579   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   580   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
   581 }
   584 void TemplateTable::daload() {
   585   transition(itos, dtos);
   586   // Otos_i: index
   587   // O2: array
   588   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   589   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
   590 }
   593 void TemplateTable::aaload() {
   594   transition(itos, atos);
   595   // Otos_i: index
   596   // tos: array
   597   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
   598   __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
   599   __ verify_oop(Otos_i);
   600 }
   603 void TemplateTable::baload() {
   604   transition(itos, itos);
   605   // Otos_i: index
   606   // tos: array
   607   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
   608   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
   609 }
   612 void TemplateTable::caload() {
   613   transition(itos, itos);
   614   // Otos_i: index
   615   // tos: array
   616   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   617   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   618 }
   620 void TemplateTable::fast_icaload() {
   621   transition(vtos, itos);
   622   // Otos_i: index
   623   // tos: array
   624   locals_index(G3_scratch);
   625   __ access_local_int( G3_scratch, Otos_i );
   626   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   627   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   628 }
   631 void TemplateTable::saload() {
   632   transition(itos, itos);
   633   // Otos_i: index
   634   // tos: array
   635   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   636   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
   637 }
   640 void TemplateTable::iload(int n) {
   641   transition(vtos, itos);
   642   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   643 }
   646 void TemplateTable::lload(int n) {
   647   transition(vtos, ltos);
   648   assert(n+1 < Argument::n_register_parameters, "would need more code");
   649   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
   650 }
   653 void TemplateTable::fload(int n) {
   654   transition(vtos, ftos);
   655   assert(n < Argument::n_register_parameters, "would need more code");
   656   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
   657 }
   660 void TemplateTable::dload(int n) {
   661   transition(vtos, dtos);
   662   FloatRegister dst = Ftos_d;
   663   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
   664 }
   667 void TemplateTable::aload(int n) {
   668   transition(vtos, atos);
   669   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   670 }
   673 void TemplateTable::aload_0() {
   674   transition(vtos, atos);
   676   // According to bytecode histograms, the pairs:
   677   //
   678   // _aload_0, _fast_igetfield (itos)
   679   // _aload_0, _fast_agetfield (atos)
   680   // _aload_0, _fast_fgetfield (ftos)
   681   //
   682   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
   683   // bytecode checks the next bytecode and then rewrites the current
   684   // bytecode into a pair bytecode; otherwise it rewrites the current
   685   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
   686   //
   687   if (RewriteFrequentPairs) {
   688     Label rewrite, done;
   690     // get next byte
   691     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
   693     // do actual aload_0
   694     aload(0);
   696     // if _getfield then wait with rewrite
   697     __ cmp(G3_scratch, (int)Bytecodes::_getfield);
   698     __ br(Assembler::equal, false, Assembler::pn, done);
   699     __ delayed()->nop();
   701     // if _igetfield then rewrite to _fast_iaccess_0
   702     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   703     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
   704     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   705     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
   707     // if _agetfield then rewrite to _fast_aaccess_0
   708     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   709     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
   710     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   711     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
   713     // if _fgetfield then rewrite to _fast_faccess_0
   714     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   715     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
   716     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   717     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
   719     // else rewrite to _fast_aload0
   720     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   721     __ set(Bytecodes::_fast_aload_0, G4_scratch);
   723     // rewrite
   724     // G4_scratch: fast bytecode
   725     __ bind(rewrite);
   726     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
   727     __ bind(done);
   728   } else {
   729     aload(0);
   730   }
   731 }
   734 void TemplateTable::istore() {
   735   transition(itos, vtos);
   736   locals_index(G3_scratch);
   737   __ store_local_int( G3_scratch, Otos_i );
   738 }
   741 void TemplateTable::lstore() {
   742   transition(ltos, vtos);
   743   locals_index(G3_scratch);
   744   __ store_local_long( G3_scratch, Otos_l );
   745 }
   748 void TemplateTable::fstore() {
   749   transition(ftos, vtos);
   750   locals_index(G3_scratch);
   751   __ store_local_float( G3_scratch, Ftos_f );
   752 }
   755 void TemplateTable::dstore() {
   756   transition(dtos, vtos);
   757   locals_index(G3_scratch);
   758   __ store_local_double( G3_scratch, Ftos_d );
   759 }
   762 void TemplateTable::astore() {
   763   transition(vtos, vtos);
   764   __ load_ptr(0, Otos_i);
   765   __ inc(Lesp, Interpreter::stackElementSize);
   766   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   767   locals_index(G3_scratch);
   768   __ store_local_ptr(G3_scratch, Otos_i);
   769 }
   772 void TemplateTable::wide_istore() {
   773   transition(vtos, vtos);
   774   __ pop_i();
   775   locals_index_wide(G3_scratch);
   776   __ store_local_int( G3_scratch, Otos_i );
   777 }
   780 void TemplateTable::wide_lstore() {
   781   transition(vtos, vtos);
   782   __ pop_l();
   783   locals_index_wide(G3_scratch);
   784   __ store_local_long( G3_scratch, Otos_l );
   785 }
   788 void TemplateTable::wide_fstore() {
   789   transition(vtos, vtos);
   790   __ pop_f();
   791   locals_index_wide(G3_scratch);
   792   __ store_local_float( G3_scratch, Ftos_f );
   793 }
   796 void TemplateTable::wide_dstore() {
   797   transition(vtos, vtos);
   798   __ pop_d();
   799   locals_index_wide(G3_scratch);
   800   __ store_local_double( G3_scratch, Ftos_d );
   801 }
   804 void TemplateTable::wide_astore() {
   805   transition(vtos, vtos);
   806   __ load_ptr(0, Otos_i);
   807   __ inc(Lesp, Interpreter::stackElementSize);
   808   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   809   locals_index_wide(G3_scratch);
   810   __ store_local_ptr(G3_scratch, Otos_i);
   811 }
   814 void TemplateTable::iastore() {
   815   transition(itos, vtos);
   816   __ pop_i(O2); // index
   817   // Otos_i: val
   818   // O3: array
   819   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   820   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
   821 }
   824 void TemplateTable::lastore() {
   825   transition(ltos, vtos);
   826   __ pop_i(O2); // index
   827   // Otos_l: val
   828   // O3: array
   829   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   830   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
   831 }
   834 void TemplateTable::fastore() {
   835   transition(ftos, vtos);
   836   __ pop_i(O2); // index
   837   // Ftos_f: val
   838   // O3: array
   839   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   840   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
   841 }
   844 void TemplateTable::dastore() {
   845   transition(dtos, vtos);
   846   __ pop_i(O2); // index
   847   // Fos_d: val
   848   // O3: array
   849   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   850   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
   851 }
   854 void TemplateTable::aastore() {
   855   Label store_ok, is_null, done;
   856   transition(vtos, vtos);
   857   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
   858   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
   859   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
   860   // Otos_i: val
   861   // O2: index
   862   // O3: array
   863   __ verify_oop(Otos_i);
   864   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
   866   // do array store check - check for NULL value first
   867   __ br_null( Otos_i, false, Assembler::pn, is_null );
   868   __ delayed()->nop();
   870   __ load_klass(O3, O4); // get array klass
   871   __ load_klass(Otos_i, O5); // get value klass
   873   // do fast instanceof cache test
   875   __ ld_ptr(O4,     sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes(),  O4);
   877   assert(Otos_i == O0, "just checking");
   879   // Otos_i:    value
   880   // O1:        addr - offset
   881   // O2:        index
   882   // O3:        array
   883   // O4:        array element klass
   884   // O5:        value klass
   886   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
   888   // Generate a fast subtype check.  Branch to store_ok if no
   889   // failure.  Throw if failure.
   890   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
   892   // Not a subtype; so must throw exception
   893   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
   895   // Store is OK.
   896   __ bind(store_ok);
   897   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
   899   __ ba(false,done);
   900   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
   902   __ bind(is_null);
   903   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
   905   __ profile_null_seen(G3_scratch);
   906   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
   907   __ bind(done);
   908 }
   911 void TemplateTable::bastore() {
   912   transition(itos, vtos);
   913   __ pop_i(O2); // index
   914   // Otos_i: val
   915   // O3: array
   916   __ index_check(O3, O2, 0, G3_scratch, O2);
   917   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
   918 }
   921 void TemplateTable::castore() {
   922   transition(itos, vtos);
   923   __ pop_i(O2); // index
   924   // Otos_i: val
   925   // O3: array
   926   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
   927   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
   928 }
   931 void TemplateTable::sastore() {
   932   // %%%%% Factor across platform
   933   castore();
   934 }
   937 void TemplateTable::istore(int n) {
   938   transition(itos, vtos);
   939   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
   940 }
   943 void TemplateTable::lstore(int n) {
   944   transition(ltos, vtos);
   945   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
   946   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
   948 }
   951 void TemplateTable::fstore(int n) {
   952   transition(ftos, vtos);
   953   assert(n < Argument::n_register_parameters, "only handle register cases");
   954   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
   955 }
   958 void TemplateTable::dstore(int n) {
   959   transition(dtos, vtos);
   960   FloatRegister src = Ftos_d;
   961   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
   962 }
   965 void TemplateTable::astore(int n) {
   966   transition(vtos, vtos);
   967   __ load_ptr(0, Otos_i);
   968   __ inc(Lesp, Interpreter::stackElementSize);
   969   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   970   __ store_local_ptr(n, Otos_i);
   971 }
   974 void TemplateTable::pop() {
   975   transition(vtos, vtos);
   976   __ inc(Lesp, Interpreter::stackElementSize);
   977 }
   980 void TemplateTable::pop2() {
   981   transition(vtos, vtos);
   982   __ inc(Lesp, 2 * Interpreter::stackElementSize);
   983 }
   986 void TemplateTable::dup() {
   987   transition(vtos, vtos);
   988   // stack: ..., a
   989   // load a and tag
   990   __ load_ptr(0, Otos_i);
   991   __ push_ptr(Otos_i);
   992   // stack: ..., a, a
   993 }
   996 void TemplateTable::dup_x1() {
   997   transition(vtos, vtos);
   998   // stack: ..., a, b
   999   __ load_ptr( 1, G3_scratch);  // get a
  1000   __ load_ptr( 0, Otos_l1);     // get b
  1001   __ store_ptr(1, Otos_l1);     // put b
  1002   __ store_ptr(0, G3_scratch);  // put a - like swap
  1003   __ push_ptr(Otos_l1);         // push b
  1004   // stack: ..., b, a, b
  1008 void TemplateTable::dup_x2() {
  1009   transition(vtos, vtos);
  1010   // stack: ..., a, b, c
  1011   // get c and push on stack, reuse registers
  1012   __ load_ptr( 0, G3_scratch);  // get c
  1013   __ push_ptr(G3_scratch);      // push c with tag
  1014   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
  1015   // (stack offsets n+1 now)
  1016   __ load_ptr( 3, Otos_l1);     // get a
  1017   __ store_ptr(3, G3_scratch);  // put c at 3
  1018   // stack: ..., c, b, c, c  (a in reg)
  1019   __ load_ptr( 2, G3_scratch);  // get b
  1020   __ store_ptr(2, Otos_l1);     // put a at 2
  1021   // stack: ..., c, a, c, c  (b in reg)
  1022   __ store_ptr(1, G3_scratch);  // put b at 1
  1023   // stack: ..., c, a, b, c
  1027 void TemplateTable::dup2() {
  1028   transition(vtos, vtos);
  1029   __ load_ptr(1, G3_scratch);  // get a
  1030   __ load_ptr(0, Otos_l1);     // get b
  1031   __ push_ptr(G3_scratch);     // push a
  1032   __ push_ptr(Otos_l1);        // push b
  1033   // stack: ..., a, b, a, b
  1037 void TemplateTable::dup2_x1() {
  1038   transition(vtos, vtos);
  1039   // stack: ..., a, b, c
  1040   __ load_ptr( 1, Lscratch);    // get b
  1041   __ load_ptr( 2, Otos_l1);     // get a
  1042   __ store_ptr(2, Lscratch);    // put b at a
  1043   // stack: ..., b, b, c
  1044   __ load_ptr( 0, G3_scratch);  // get c
  1045   __ store_ptr(1, G3_scratch);  // put c at b
  1046   // stack: ..., b, c, c
  1047   __ store_ptr(0, Otos_l1);     // put a at c
  1048   // stack: ..., b, c, a
  1049   __ push_ptr(Lscratch);        // push b
  1050   __ push_ptr(G3_scratch);      // push c
  1051   // stack: ..., b, c, a, b, c
  1055 // The spec says that these types can be a mixture of category 1 (1 word)
  1056 // types and/or category 2 types (long and doubles)
  1057 void TemplateTable::dup2_x2() {
  1058   transition(vtos, vtos);
  1059   // stack: ..., a, b, c, d
  1060   __ load_ptr( 1, Lscratch);    // get c
  1061   __ load_ptr( 3, Otos_l1);     // get a
  1062   __ store_ptr(3, Lscratch);    // put c at 3
  1063   __ store_ptr(1, Otos_l1);     // put a at 1
  1064   // stack: ..., c, b, a, d
  1065   __ load_ptr( 2, G3_scratch);  // get b
  1066   __ load_ptr( 0, Otos_l1);     // get d
  1067   __ store_ptr(0, G3_scratch);  // put b at 0
  1068   __ store_ptr(2, Otos_l1);     // put d at 2
  1069   // stack: ..., c, d, a, b
  1070   __ push_ptr(Lscratch);        // push c
  1071   __ push_ptr(Otos_l1);         // push d
  1072   // stack: ..., c, d, a, b, c, d
  1076 void TemplateTable::swap() {
  1077   transition(vtos, vtos);
  1078   // stack: ..., a, b
  1079   __ load_ptr( 1, G3_scratch);  // get a
  1080   __ load_ptr( 0, Otos_l1);     // get b
  1081   __ store_ptr(0, G3_scratch);  // put b
  1082   __ store_ptr(1, Otos_l1);     // put a
  1083   // stack: ..., b, a
  1087 void TemplateTable::iop2(Operation op) {
  1088   transition(itos, itos);
  1089   __ pop_i(O1);
  1090   switch (op) {
  1091    case  add:  __  add(O1, Otos_i, Otos_i);  break;
  1092    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
  1093      // %%%%% Mul may not exist: better to call .mul?
  1094    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
  1095    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
  1096    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
  1097    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
  1098    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
  1099    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
  1100    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
  1101    default: ShouldNotReachHere();
  1106 void TemplateTable::lop2(Operation op) {
  1107   transition(ltos, ltos);
  1108   __ pop_l(O2);
  1109   switch (op) {
  1110 #ifdef _LP64
  1111    case  add:  __  add(O2, Otos_l, Otos_l);  break;
  1112    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
  1113    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
  1114    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
  1115    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
  1116 #else
  1117    case  add:  __ addcc(O3, Otos_l2, Otos_l2);  __ addc(O2, Otos_l1, Otos_l1);  break;
  1118    case  sub:  __ subcc(O3, Otos_l2, Otos_l2);  __ subc(O2, Otos_l1, Otos_l1);  break;
  1119    case _and:  __  and3(O3, Otos_l2, Otos_l2);  __ and3(O2, Otos_l1, Otos_l1);  break;
  1120    case  _or:  __   or3(O3, Otos_l2, Otos_l2);  __  or3(O2, Otos_l1, Otos_l1);  break;
  1121    case _xor:  __  xor3(O3, Otos_l2, Otos_l2);  __ xor3(O2, Otos_l1, Otos_l1);  break;
  1122 #endif
  1123    default: ShouldNotReachHere();
  1128 void TemplateTable::idiv() {
  1129   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
  1130   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
  1132   transition(itos, itos);
  1133   __ pop_i(O1); // get 1st op
  1135   // Y contains upper 32 bits of result, set it to 0 or all ones
  1136   __ wry(G0);
  1137   __ mov(~0, G3_scratch);
  1139   __ tst(O1);
  1140      Label neg;
  1141   __ br(Assembler::negative, true, Assembler::pn, neg);
  1142   __ delayed()->wry(G3_scratch);
  1143   __ bind(neg);
  1145      Label ok;
  1146   __ tst(Otos_i);
  1147   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
  1149   const int min_int = 0x80000000;
  1150   Label regular;
  1151   __ cmp(Otos_i, -1);
  1152   __ br(Assembler::notEqual, false, Assembler::pt, regular);
  1153 #ifdef _LP64
  1154   // Don't put set in delay slot
  1155   // Set will turn into multiple instructions in 64 bit mode
  1156   __ delayed()->nop();
  1157   __ set(min_int, G4_scratch);
  1158 #else
  1159   __ delayed()->set(min_int, G4_scratch);
  1160 #endif
  1161   Label done;
  1162   __ cmp(O1, G4_scratch);
  1163   __ br(Assembler::equal, true, Assembler::pt, done);
  1164   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
  1166   __ bind(regular);
  1167   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
  1168   __ bind(done);
  1172 void TemplateTable::irem() {
  1173   transition(itos, itos);
  1174   __ mov(Otos_i, O2); // save divisor
  1175   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
  1176   __ smul(Otos_i, O2, Otos_i);
  1177   __ sub(O1, Otos_i, Otos_i);
  1181 void TemplateTable::lmul() {
  1182   transition(ltos, ltos);
  1183   __ pop_l(O2);
  1184 #ifdef _LP64
  1185   __ mulx(Otos_l, O2, Otos_l);
  1186 #else
  1187   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lmul));
  1188 #endif
  1193 void TemplateTable::ldiv() {
  1194   transition(ltos, ltos);
  1196   // check for zero
  1197   __ pop_l(O2);
  1198 #ifdef _LP64
  1199   __ tst(Otos_l);
  1200   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1201   __ sdivx(O2, Otos_l, Otos_l);
  1202 #else
  1203   __ orcc(Otos_l1, Otos_l2, G0);
  1204   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1205   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
  1206 #endif
  1210 void TemplateTable::lrem() {
  1211   transition(ltos, ltos);
  1213   // check for zero
  1214   __ pop_l(O2);
  1215 #ifdef _LP64
  1216   __ tst(Otos_l);
  1217   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1218   __ sdivx(O2, Otos_l, Otos_l2);
  1219   __ mulx (Otos_l2, Otos_l, Otos_l2);
  1220   __ sub  (O2, Otos_l2, Otos_l);
  1221 #else
  1222   __ orcc(Otos_l1, Otos_l2, G0);
  1223   __ throw_if_not_icc(Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1224   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
  1225 #endif
  1229 void TemplateTable::lshl() {
  1230   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
  1232   __ pop_l(O2);                          // shift value in O2, O3
  1233 #ifdef _LP64
  1234   __ sllx(O2, Otos_i, Otos_l);
  1235 #else
  1236   __ lshl(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1237 #endif
  1241 void TemplateTable::lshr() {
  1242   transition(itos, ltos); // %%%% see lshl comment
  1244   __ pop_l(O2);                          // shift value in O2, O3
  1245 #ifdef _LP64
  1246   __ srax(O2, Otos_i, Otos_l);
  1247 #else
  1248   __ lshr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1249 #endif
  1254 void TemplateTable::lushr() {
  1255   transition(itos, ltos); // %%%% see lshl comment
  1257   __ pop_l(O2);                          // shift value in O2, O3
  1258 #ifdef _LP64
  1259   __ srlx(O2, Otos_i, Otos_l);
  1260 #else
  1261   __ lushr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1262 #endif
  1266 void TemplateTable::fop2(Operation op) {
  1267   transition(ftos, ftos);
  1268   switch (op) {
  1269    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1270    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1271    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1272    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1273    case  rem:
  1274      assert(Ftos_f == F0, "just checking");
  1275 #ifdef _LP64
  1276      // LP64 calling conventions use F1, F3 for passing 2 floats
  1277      __ pop_f(F1);
  1278      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
  1279 #else
  1280      __ pop_i(O0);
  1281      __ stf(FloatRegisterImpl::S, Ftos_f, __ d_tmp);
  1282      __ ld( __ d_tmp, O1 );
  1283 #endif
  1284      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
  1285      assert( Ftos_f == F0, "fix this code" );
  1286      break;
  1288    default: ShouldNotReachHere();
  1293 void TemplateTable::dop2(Operation op) {
  1294   transition(dtos, dtos);
  1295   switch (op) {
  1296    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1297    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1298    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1299    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1300    case  rem:
  1301 #ifdef _LP64
  1302      // Pass arguments in D0, D2
  1303      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
  1304      __ pop_d( F0 );
  1305 #else
  1306      // Pass arguments in O0O1, O2O3
  1307      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1308      __ ldd( __ d_tmp, O2 );
  1309      __ pop_d(Ftos_f);
  1310      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1311      __ ldd( __ d_tmp, O0 );
  1312 #endif
  1313      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
  1314      assert( Ftos_d == F0, "fix this code" );
  1315      break;
  1317    default: ShouldNotReachHere();
  1322 void TemplateTable::ineg() {
  1323   transition(itos, itos);
  1324   __ neg(Otos_i);
  1328 void TemplateTable::lneg() {
  1329   transition(ltos, ltos);
  1330 #ifdef _LP64
  1331   __ sub(G0, Otos_l, Otos_l);
  1332 #else
  1333   __ lneg(Otos_l1, Otos_l2);
  1334 #endif
  1338 void TemplateTable::fneg() {
  1339   transition(ftos, ftos);
  1340   __ fneg(FloatRegisterImpl::S, Ftos_f);
  1344 void TemplateTable::dneg() {
  1345   transition(dtos, dtos);
  1346   // v8 has fnegd if source and dest are the same
  1347   __ fneg(FloatRegisterImpl::D, Ftos_f);
  1351 void TemplateTable::iinc() {
  1352   transition(vtos, vtos);
  1353   locals_index(G3_scratch);
  1354   __ ldsb(Lbcp, 2, O2);  // load constant
  1355   __ access_local_int(G3_scratch, Otos_i);
  1356   __ add(Otos_i, O2, Otos_i);
  1357   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1361 void TemplateTable::wide_iinc() {
  1362   transition(vtos, vtos);
  1363   locals_index_wide(G3_scratch);
  1364   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
  1365   __ access_local_int(G3_scratch, Otos_i);
  1366   __ add(Otos_i, O3, Otos_i);
  1367   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1371 void TemplateTable::convert() {
  1372 // %%%%% Factor this first part accross platforms
  1373   #ifdef ASSERT
  1374     TosState tos_in  = ilgl;
  1375     TosState tos_out = ilgl;
  1376     switch (bytecode()) {
  1377       case Bytecodes::_i2l: // fall through
  1378       case Bytecodes::_i2f: // fall through
  1379       case Bytecodes::_i2d: // fall through
  1380       case Bytecodes::_i2b: // fall through
  1381       case Bytecodes::_i2c: // fall through
  1382       case Bytecodes::_i2s: tos_in = itos; break;
  1383       case Bytecodes::_l2i: // fall through
  1384       case Bytecodes::_l2f: // fall through
  1385       case Bytecodes::_l2d: tos_in = ltos; break;
  1386       case Bytecodes::_f2i: // fall through
  1387       case Bytecodes::_f2l: // fall through
  1388       case Bytecodes::_f2d: tos_in = ftos; break;
  1389       case Bytecodes::_d2i: // fall through
  1390       case Bytecodes::_d2l: // fall through
  1391       case Bytecodes::_d2f: tos_in = dtos; break;
  1392       default             : ShouldNotReachHere();
  1394     switch (bytecode()) {
  1395       case Bytecodes::_l2i: // fall through
  1396       case Bytecodes::_f2i: // fall through
  1397       case Bytecodes::_d2i: // fall through
  1398       case Bytecodes::_i2b: // fall through
  1399       case Bytecodes::_i2c: // fall through
  1400       case Bytecodes::_i2s: tos_out = itos; break;
  1401       case Bytecodes::_i2l: // fall through
  1402       case Bytecodes::_f2l: // fall through
  1403       case Bytecodes::_d2l: tos_out = ltos; break;
  1404       case Bytecodes::_i2f: // fall through
  1405       case Bytecodes::_l2f: // fall through
  1406       case Bytecodes::_d2f: tos_out = ftos; break;
  1407       case Bytecodes::_i2d: // fall through
  1408       case Bytecodes::_l2d: // fall through
  1409       case Bytecodes::_f2d: tos_out = dtos; break;
  1410       default             : ShouldNotReachHere();
  1412     transition(tos_in, tos_out);
  1413   #endif
  1416   // Conversion
  1417   Label done;
  1418   switch (bytecode()) {
  1419    case Bytecodes::_i2l:
  1420 #ifdef _LP64
  1421     // Sign extend the 32 bits
  1422     __ sra ( Otos_i, 0, Otos_l );
  1423 #else
  1424     __ addcc(Otos_i, 0, Otos_l2);
  1425     __ br(Assembler::greaterEqual, true, Assembler::pt, done);
  1426     __ delayed()->clr(Otos_l1);
  1427     __ set(~0, Otos_l1);
  1428 #endif
  1429     break;
  1431    case Bytecodes::_i2f:
  1432     __ st(Otos_i, __ d_tmp );
  1433     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1434     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
  1435     break;
  1437    case Bytecodes::_i2d:
  1438     __ st(Otos_i, __ d_tmp);
  1439     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1440     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
  1441     break;
  1443    case Bytecodes::_i2b:
  1444     __ sll(Otos_i, 24, Otos_i);
  1445     __ sra(Otos_i, 24, Otos_i);
  1446     break;
  1448    case Bytecodes::_i2c:
  1449     __ sll(Otos_i, 16, Otos_i);
  1450     __ srl(Otos_i, 16, Otos_i);
  1451     break;
  1453    case Bytecodes::_i2s:
  1454     __ sll(Otos_i, 16, Otos_i);
  1455     __ sra(Otos_i, 16, Otos_i);
  1456     break;
  1458    case Bytecodes::_l2i:
  1459 #ifndef _LP64
  1460     __ mov(Otos_l2, Otos_i);
  1461 #else
  1462     // Sign-extend into the high 32 bits
  1463     __ sra(Otos_l, 0, Otos_i);
  1464 #endif
  1465     break;
  1467    case Bytecodes::_l2f:
  1468    case Bytecodes::_l2d:
  1469     __ st_long(Otos_l, __ d_tmp);
  1470     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
  1472     if (VM_Version::v9_instructions_work()) {
  1473       if (bytecode() == Bytecodes::_l2f) {
  1474         __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1475       } else {
  1476         __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
  1478     } else {
  1479       __ call_VM_leaf(
  1480         Lscratch,
  1481         bytecode() == Bytecodes::_l2f
  1482           ? CAST_FROM_FN_PTR(address, SharedRuntime::l2f)
  1483           : CAST_FROM_FN_PTR(address, SharedRuntime::l2d)
  1484       );
  1486     break;
  1488   case Bytecodes::_f2i:  {
  1489       Label isNaN;
  1490       // result must be 0 if value is NaN; test by comparing value to itself
  1491       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
  1492       // According to the v8 manual, you have to have a non-fp instruction
  1493       // between fcmp and fb.
  1494       if (!VM_Version::v9_instructions_work()) {
  1495         __ nop();
  1497       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
  1498       __ delayed()->clr(Otos_i);                                     // NaN
  1499       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
  1500       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
  1501       __ ld(__ d_tmp, Otos_i);
  1502       __ bind(isNaN);
  1504     break;
  1506    case Bytecodes::_f2l:
  1507     // must uncache tos
  1508     __ push_f();
  1509 #ifdef _LP64
  1510     __ pop_f(F1);
  1511 #else
  1512     __ pop_i(O0);
  1513 #endif
  1514     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
  1515     break;
  1517    case Bytecodes::_f2d:
  1518     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
  1519     break;
  1521    case Bytecodes::_d2i:
  1522    case Bytecodes::_d2l:
  1523     // must uncache tos
  1524     __ push_d();
  1525 #ifdef _LP64
  1526     // LP64 calling conventions pass first double arg in D0
  1527     __ pop_d( Ftos_d );
  1528 #else
  1529     __ pop_i( O0 );
  1530     __ pop_i( O1 );
  1531 #endif
  1532     __ call_VM_leaf(Lscratch,
  1533         bytecode() == Bytecodes::_d2i
  1534           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
  1535           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
  1536     break;
  1538     case Bytecodes::_d2f:
  1539     if (VM_Version::v9_instructions_work()) {
  1540       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1542     else {
  1543       // must uncache tos
  1544       __ push_d();
  1545       __ pop_i(O0);
  1546       __ pop_i(O1);
  1547       __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::d2f));
  1549     break;
  1551     default: ShouldNotReachHere();
  1553   __ bind(done);
  1557 void TemplateTable::lcmp() {
  1558   transition(ltos, itos);
  1560 #ifdef _LP64
  1561   __ pop_l(O1); // pop off value 1, value 2 is in O0
  1562   __ lcmp( O1, Otos_l, Otos_i );
  1563 #else
  1564   __ pop_l(O2); // cmp O2,3 to O0,1
  1565   __ lcmp( O2, O3, Otos_l1, Otos_l2, Otos_i );
  1566 #endif
  1570 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
  1572   if (is_float) __ pop_f(F2);
  1573   else          __ pop_d(F2);
  1575   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
  1577   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
  1580 void TemplateTable::branch(bool is_jsr, bool is_wide) {
  1581   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
  1582   __ verify_oop(Lmethod);
  1583   __ verify_thread();
  1585   const Register O2_bumped_count = O2;
  1586   __ profile_taken_branch(G3_scratch, O2_bumped_count);
  1588   // get (wide) offset to O1_disp
  1589   const Register O1_disp = O1;
  1590   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
  1591   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
  1593   // Handle all the JSR stuff here, then exit.
  1594   // It's much shorter and cleaner than intermingling with the
  1595   // non-JSR normal-branch stuff occurring below.
  1596   if( is_jsr ) {
  1597     // compute return address as bci in Otos_i
  1598     __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1599     __ sub(Lbcp, G3_scratch, G3_scratch);
  1600     __ sub(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
  1602     // Bump Lbcp to target of JSR
  1603     __ add(Lbcp, O1_disp, Lbcp);
  1604     // Push returnAddress for "ret" on stack
  1605     __ push_ptr(Otos_i);
  1606     // And away we go!
  1607     __ dispatch_next(vtos);
  1608     return;
  1611   // Normal (non-jsr) branch handling
  1613   // Save the current Lbcp
  1614   const Register O0_cur_bcp = O0;
  1615   __ mov( Lbcp, O0_cur_bcp );
  1618   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
  1619   if ( increment_invocation_counter_for_backward_branches ) {
  1620     Label Lforward;
  1621     // check branch direction
  1622     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
  1623     // Bump bytecode pointer by displacement (take the branch)
  1624     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
  1626     if (TieredCompilation) {
  1627       Label Lno_mdo, Loverflow;
  1628       int increment = InvocationCounter::count_increment;
  1629       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
  1630       if (ProfileInterpreter) {
  1631         // If no method data exists, go to profile_continue.
  1632         __ ld_ptr(Lmethod, methodOopDesc::method_data_offset(), G4_scratch);
  1633         __ br_null(G4_scratch, false, Assembler::pn, Lno_mdo);
  1634         __ delayed()->nop();
  1636         // Increment backedge counter in the MDO
  1637         Address mdo_backedge_counter(G4_scratch, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
  1638                                                  in_bytes(InvocationCounter::counter_offset()));
  1639         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, Lscratch,
  1640                                    Assembler::notZero, &Lforward);
  1641         __ ba(false, Loverflow);
  1642         __ delayed()->nop();
  1645       // If there's no MDO, increment counter in methodOop
  1646       __ bind(Lno_mdo);
  1647       Address backedge_counter(Lmethod, in_bytes(methodOopDesc::backedge_counter_offset()) +
  1648                                         in_bytes(InvocationCounter::counter_offset()));
  1649       __ increment_mask_and_jump(backedge_counter, increment, mask, G3_scratch, Lscratch,
  1650                                  Assembler::notZero, &Lforward);
  1651       __ bind(Loverflow);
  1653       // notify point for loop, pass branch bytecode
  1654       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), O0_cur_bcp);
  1656       // Was an OSR adapter generated?
  1657       // O0 = osr nmethod
  1658       __ br_null(O0, false, Assembler::pn, Lforward);
  1659       __ delayed()->nop();
  1661       // Has the nmethod been invalidated already?
  1662       __ ld(O0, nmethod::entry_bci_offset(), O2);
  1663       __ cmp(O2, InvalidOSREntryBci);
  1664       __ br(Assembler::equal, false, Assembler::pn, Lforward);
  1665       __ delayed()->nop();
  1667       // migrate the interpreter frame off of the stack
  1669       __ mov(G2_thread, L7);
  1670       // save nmethod
  1671       __ mov(O0, L6);
  1672       __ set_last_Java_frame(SP, noreg);
  1673       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
  1674       __ reset_last_Java_frame();
  1675       __ mov(L7, G2_thread);
  1677       // move OSR nmethod to I1
  1678       __ mov(L6, I1);
  1680       // OSR buffer to I0
  1681       __ mov(O0, I0);
  1683       // remove the interpreter frame
  1684       __ restore(I5_savedSP, 0, SP);
  1686       // Jump to the osr code.
  1687       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
  1688       __ jmp(O2, G0);
  1689       __ delayed()->nop();
  1691     } else {
  1692       // Update Backedge branch separately from invocations
  1693       const Register G4_invoke_ctr = G4;
  1694       __ increment_backedge_counter(G4_invoke_ctr, G1_scratch);
  1695       if (ProfileInterpreter) {
  1696         __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_scratch, Lforward);
  1697         if (UseOnStackReplacement) {
  1698           __ test_backedge_count_for_osr(O2_bumped_count, O0_cur_bcp, G3_scratch);
  1700       } else {
  1701         if (UseOnStackReplacement) {
  1702           __ test_backedge_count_for_osr(G4_invoke_ctr, O0_cur_bcp, G3_scratch);
  1707     __ bind(Lforward);
  1708   } else
  1709     // Bump bytecode pointer by displacement (take the branch)
  1710     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
  1712   // continue with bytecode @ target
  1713   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
  1714   // %%%%% and changing dispatch_next to dispatch_only
  1715   __ dispatch_next(vtos);
  1719 // Note Condition in argument is TemplateTable::Condition
  1720 // arg scope is within class scope
  1722 void TemplateTable::if_0cmp(Condition cc) {
  1723   // no pointers, integer only!
  1724   transition(itos, vtos);
  1725   // assume branch is more often taken than not (loops use backward branches)
  1726   __ cmp( Otos_i, 0);
  1727   __ if_cmp(ccNot(cc), false);
  1731 void TemplateTable::if_icmp(Condition cc) {
  1732   transition(itos, vtos);
  1733   __ pop_i(O1);
  1734   __ cmp(O1, Otos_i);
  1735   __ if_cmp(ccNot(cc), false);
  1739 void TemplateTable::if_nullcmp(Condition cc) {
  1740   transition(atos, vtos);
  1741   __ tst(Otos_i);
  1742   __ if_cmp(ccNot(cc), true);
  1746 void TemplateTable::if_acmp(Condition cc) {
  1747   transition(atos, vtos);
  1748   __ pop_ptr(O1);
  1749   __ verify_oop(O1);
  1750   __ verify_oop(Otos_i);
  1751   __ cmp(O1, Otos_i);
  1752   __ if_cmp(ccNot(cc), true);
  1757 void TemplateTable::ret() {
  1758   transition(vtos, vtos);
  1759   locals_index(G3_scratch);
  1760   __ access_local_returnAddress(G3_scratch, Otos_i);
  1761   // Otos_i contains the bci, compute the bcp from that
  1763 #ifdef _LP64
  1764 #ifdef ASSERT
  1765   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
  1766   // the result.  The return address (really a BCI) was stored with an
  1767   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
  1768   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
  1769   // loaded value.
  1770   { Label zzz ;
  1771      __ set (65536, G3_scratch) ;
  1772      __ cmp (Otos_i, G3_scratch) ;
  1773      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
  1774      __ delayed()->nop();
  1775      __ stop("BCI is in the wrong register half?");
  1776      __ bind (zzz) ;
  1778 #endif
  1779 #endif
  1781   __ profile_ret(vtos, Otos_i, G4_scratch);
  1783   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1784   __ add(G3_scratch, Otos_i, G3_scratch);
  1785   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
  1786   __ dispatch_next(vtos);
  1790 void TemplateTable::wide_ret() {
  1791   transition(vtos, vtos);
  1792   locals_index_wide(G3_scratch);
  1793   __ access_local_returnAddress(G3_scratch, Otos_i);
  1794   // Otos_i contains the bci, compute the bcp from that
  1796   __ profile_ret(vtos, Otos_i, G4_scratch);
  1798   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1799   __ add(G3_scratch, Otos_i, G3_scratch);
  1800   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
  1801   __ dispatch_next(vtos);
  1805 void TemplateTable::tableswitch() {
  1806   transition(itos, vtos);
  1807   Label default_case, continue_execution;
  1809   // align bcp
  1810   __ add(Lbcp, BytesPerInt, O1);
  1811   __ and3(O1, -BytesPerInt, O1);
  1812   // load lo, hi
  1813   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
  1814   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
  1815 #ifdef _LP64
  1816   // Sign extend the 32 bits
  1817   __ sra ( Otos_i, 0, Otos_i );
  1818 #endif /* _LP64 */
  1820   // check against lo & hi
  1821   __ cmp( Otos_i, O2);
  1822   __ br( Assembler::less, false, Assembler::pn, default_case);
  1823   __ delayed()->cmp( Otos_i, O3 );
  1824   __ br( Assembler::greater, false, Assembler::pn, default_case);
  1825   // lookup dispatch offset
  1826   __ delayed()->sub(Otos_i, O2, O2);
  1827   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
  1828   __ sll(O2, LogBytesPerInt, O2);
  1829   __ add(O2, 3 * BytesPerInt, O2);
  1830   __ ba(false, continue_execution);
  1831   __ delayed()->ld(O1, O2, O2);
  1832   // handle default
  1833   __ bind(default_case);
  1834   __ profile_switch_default(O3);
  1835   __ ld(O1, 0, O2); // get default offset
  1836   // continue execution
  1837   __ bind(continue_execution);
  1838   __ add(Lbcp, O2, Lbcp);
  1839   __ dispatch_next(vtos);
  1843 void TemplateTable::lookupswitch() {
  1844   transition(itos, itos);
  1845   __ stop("lookupswitch bytecode should have been rewritten");
  1848 void TemplateTable::fast_linearswitch() {
  1849   transition(itos, vtos);
  1850     Label loop_entry, loop, found, continue_execution;
  1851   // align bcp
  1852   __ add(Lbcp, BytesPerInt, O1);
  1853   __ and3(O1, -BytesPerInt, O1);
  1854  // set counter
  1855   __ ld(O1, BytesPerInt, O2);
  1856   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
  1857   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
  1858   __ ba(false, loop_entry);
  1859   __ delayed()->add(O3, O2, O2); // counter now points past last pair
  1861   // table search
  1862   __ bind(loop);
  1863   __ cmp(O4, Otos_i);
  1864   __ br(Assembler::equal, true, Assembler::pn, found);
  1865   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
  1866   __ inc(O3, 2 * BytesPerInt);
  1868   __ bind(loop_entry);
  1869   __ cmp(O2, O3);
  1870   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
  1871   __ delayed()->ld(O3, 0, O4);
  1873   // default case
  1874   __ ld(O1, 0, O4); // get default offset
  1875   if (ProfileInterpreter) {
  1876     __ profile_switch_default(O3);
  1877     __ ba(false, continue_execution);
  1878     __ delayed()->nop();
  1881   // entry found -> get offset
  1882   __ bind(found);
  1883   if (ProfileInterpreter) {
  1884     __ sub(O3, O1, O3);
  1885     __ sub(O3, 2*BytesPerInt, O3);
  1886     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
  1887     __ profile_switch_case(O3, O1, O2, G3_scratch);
  1889     __ bind(continue_execution);
  1891   __ add(Lbcp, O4, Lbcp);
  1892   __ dispatch_next(vtos);
  1896 void TemplateTable::fast_binaryswitch() {
  1897   transition(itos, vtos);
  1898   // Implementation using the following core algorithm: (copied from Intel)
  1899   //
  1900   // int binary_search(int key, LookupswitchPair* array, int n) {
  1901   //   // Binary search according to "Methodik des Programmierens" by
  1902   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
  1903   //   int i = 0;
  1904   //   int j = n;
  1905   //   while (i+1 < j) {
  1906   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
  1907   //     // with      Q: for all i: 0 <= i < n: key < a[i]
  1908   //     // where a stands for the array and assuming that the (inexisting)
  1909   //     // element a[n] is infinitely big.
  1910   //     int h = (i + j) >> 1;
  1911   //     // i < h < j
  1912   //     if (key < array[h].fast_match()) {
  1913   //       j = h;
  1914   //     } else {
  1915   //       i = h;
  1916   //     }
  1917   //   }
  1918   //   // R: a[i] <= key < a[i+1] or Q
  1919   //   // (i.e., if key is within array, i is the correct index)
  1920   //   return i;
  1921   // }
  1923   // register allocation
  1924   assert(Otos_i == O0, "alias checking");
  1925   const Register Rkey     = Otos_i;                    // already set (tosca)
  1926   const Register Rarray   = O1;
  1927   const Register Ri       = O2;
  1928   const Register Rj       = O3;
  1929   const Register Rh       = O4;
  1930   const Register Rscratch = O5;
  1932   const int log_entry_size = 3;
  1933   const int entry_size = 1 << log_entry_size;
  1935   Label found;
  1936   // Find Array start
  1937   __ add(Lbcp, 3 * BytesPerInt, Rarray);
  1938   __ and3(Rarray, -BytesPerInt, Rarray);
  1939   // initialize i & j (in delay slot)
  1940   __ clr( Ri );
  1942   // and start
  1943   Label entry;
  1944   __ ba(false, entry);
  1945   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
  1946   // (Rj is already in the native byte-ordering.)
  1948   // binary search loop
  1949   { Label loop;
  1950     __ bind( loop );
  1951     // int h = (i + j) >> 1;
  1952     __ sra( Rh, 1, Rh );
  1953     // if (key < array[h].fast_match()) {
  1954     //   j = h;
  1955     // } else {
  1956     //   i = h;
  1957     // }
  1958     __ sll( Rh, log_entry_size, Rscratch );
  1959     __ ld( Rarray, Rscratch, Rscratch );
  1960     // (Rscratch is already in the native byte-ordering.)
  1961     __ cmp( Rkey, Rscratch );
  1962     if ( VM_Version::v9_instructions_work() ) {
  1963       __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
  1964       __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
  1966     else {
  1967       Label end_of_if;
  1968       __ br( Assembler::less, true, Assembler::pt, end_of_if );
  1969       __ delayed()->mov( Rh, Rj ); // if (<) Rj = Rh
  1970       __ mov( Rh, Ri );            // else i = h
  1971       __ bind(end_of_if);          // }
  1974     // while (i+1 < j)
  1975     __ bind( entry );
  1976     __ add( Ri, 1, Rscratch );
  1977     __ cmp(Rscratch, Rj);
  1978     __ br( Assembler::less, true, Assembler::pt, loop );
  1979     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
  1982   // end of binary search, result index is i (must check again!)
  1983   Label default_case;
  1984   Label continue_execution;
  1985   if (ProfileInterpreter) {
  1986     __ mov( Ri, Rh );              // Save index in i for profiling
  1988   __ sll( Ri, log_entry_size, Ri );
  1989   __ ld( Rarray, Ri, Rscratch );
  1990   // (Rscratch is already in the native byte-ordering.)
  1991   __ cmp( Rkey, Rscratch );
  1992   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
  1993   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
  1995   // entry found -> j = offset
  1996   __ inc( Ri, BytesPerInt );
  1997   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
  1998   __ ld( Rarray, Ri, Rj );
  1999   // (Rj is already in the native byte-ordering.)
  2001   if (ProfileInterpreter) {
  2002     __ ba(false, continue_execution);
  2003     __ delayed()->nop();
  2006   __ bind(default_case); // fall through (if not profiling)
  2007   __ profile_switch_default(Ri);
  2009   __ bind(continue_execution);
  2010   __ add( Lbcp, Rj, Lbcp );
  2011   __ dispatch_next( vtos );
  2015 void TemplateTable::_return(TosState state) {
  2016   transition(state, state);
  2017   assert(_desc->calls_vm(), "inconsistent calls_vm information");
  2019   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
  2020     assert(state == vtos, "only valid state");
  2021     __ mov(G0, G3_scratch);
  2022     __ access_local_ptr(G3_scratch, Otos_i);
  2023     __ load_klass(Otos_i, O2);
  2024     __ set(JVM_ACC_HAS_FINALIZER, G3);
  2025     __ ld(O2, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), O2);
  2026     __ andcc(G3, O2, G0);
  2027     Label skip_register_finalizer;
  2028     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
  2029     __ delayed()->nop();
  2031     // Call out to do finalizer registration
  2032     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
  2034     __ bind(skip_register_finalizer);
  2037   __ remove_activation(state, /* throw_monitor_exception */ true);
  2039   // The caller's SP was adjusted upon method entry to accomodate
  2040   // the callee's non-argument locals. Undo that adjustment.
  2041   __ ret();                             // return to caller
  2042   __ delayed()->restore(I5_savedSP, G0, SP);
  2046 // ----------------------------------------------------------------------------
  2047 // Volatile variables demand their effects be made known to all CPU's in
  2048 // order.  Store buffers on most chips allow reads & writes to reorder; the
  2049 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
  2050 // memory barrier (i.e., it's not sufficient that the interpreter does not
  2051 // reorder volatile references, the hardware also must not reorder them).
  2052 //
  2053 // According to the new Java Memory Model (JMM):
  2054 // (1) All volatiles are serialized wrt to each other.
  2055 // ALSO reads & writes act as aquire & release, so:
  2056 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
  2057 // the read float up to before the read.  It's OK for non-volatile memory refs
  2058 // that happen before the volatile read to float down below it.
  2059 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
  2060 // that happen BEFORE the write float down to after the write.  It's OK for
  2061 // non-volatile memory refs that happen after the volatile write to float up
  2062 // before it.
  2063 //
  2064 // We only put in barriers around volatile refs (they are expensive), not
  2065 // _between_ memory refs (that would require us to track the flavor of the
  2066 // previous memory refs).  Requirements (2) and (3) require some barriers
  2067 // before volatile stores and after volatile loads.  These nearly cover
  2068 // requirement (1) but miss the volatile-store-volatile-load case.  This final
  2069 // case is placed after volatile-stores although it could just as well go
  2070 // before volatile-loads.
  2071 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
  2072   // Helper function to insert a is-volatile test and memory barrier
  2073   // All current sparc implementations run in TSO, needing only StoreLoad
  2074   if ((order_constraint & Assembler::StoreLoad) == 0) return;
  2075   __ membar( order_constraint );
  2078 // ----------------------------------------------------------------------------
  2079 void TemplateTable::resolve_cache_and_index(int byte_no,
  2080                                             Register result,
  2081                                             Register Rcache,
  2082                                             Register index,
  2083                                             size_t index_size) {
  2084   // Depends on cpCacheOop layout!
  2085   Label resolved;
  2087   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
  2088   if (byte_no == f1_oop) {
  2089     // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
  2090     // This kind of CP cache entry does not need to match the flags byte, because
  2091     // there is a 1-1 relation between bytecode type and CP entry type.
  2092     assert_different_registers(result, Rcache);
  2093     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2094               ConstantPoolCacheEntry::f1_offset(), result);
  2095     __ tst(result);
  2096     __ br(Assembler::notEqual, false, Assembler::pt, resolved);
  2097     __ delayed()->set((int)bytecode(), O1);
  2098   } else {
  2099     assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
  2100     assert(result == noreg, "");  //else change code for setting result
  2101     const int shift_count = (1 + byte_no)*BitsPerByte;
  2103     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2104               ConstantPoolCacheEntry::indices_offset(), Lbyte_code);
  2106     __ srl(  Lbyte_code, shift_count, Lbyte_code );
  2107     __ and3( Lbyte_code,        0xFF, Lbyte_code );
  2108     __ cmp(  Lbyte_code, (int)bytecode());
  2109     __ br(   Assembler::equal, false, Assembler::pt, resolved);
  2110     __ delayed()->set((int)bytecode(), O1);
  2113   address entry;
  2114   switch (bytecode()) {
  2115     case Bytecodes::_getstatic      : // fall through
  2116     case Bytecodes::_putstatic      : // fall through
  2117     case Bytecodes::_getfield       : // fall through
  2118     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
  2119     case Bytecodes::_invokevirtual  : // fall through
  2120     case Bytecodes::_invokespecial  : // fall through
  2121     case Bytecodes::_invokestatic   : // fall through
  2122     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
  2123     case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
  2124     case Bytecodes::_fast_aldc      : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
  2125     case Bytecodes::_fast_aldc_w    : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
  2126     default                         : ShouldNotReachHere();                                 break;
  2128   // first time invocation - must resolve first
  2129   __ call_VM(noreg, entry, O1);
  2130   // Update registers with resolved info
  2131   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
  2132   if (result != noreg)
  2133     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2134               ConstantPoolCacheEntry::f1_offset(), result);
  2135   __ bind(resolved);
  2138 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
  2139                                                Register Rmethod,
  2140                                                Register Ritable_index,
  2141                                                Register Rflags,
  2142                                                bool is_invokevirtual,
  2143                                                bool is_invokevfinal,
  2144                                                bool is_invokedynamic) {
  2145   // Uses both G3_scratch and G4_scratch
  2146   Register Rcache = G3_scratch;
  2147   Register Rscratch = G4_scratch;
  2148   assert_different_registers(Rcache, Rmethod, Ritable_index);
  2150   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2152   // determine constant pool cache field offsets
  2153   const int method_offset = in_bytes(
  2154     cp_base_offset +
  2155       (is_invokevirtual
  2156        ? ConstantPoolCacheEntry::f2_offset()
  2157        : ConstantPoolCacheEntry::f1_offset()
  2159     );
  2160   const int flags_offset = in_bytes(cp_base_offset +
  2161                                     ConstantPoolCacheEntry::flags_offset());
  2162   // access constant pool cache fields
  2163   const int index_offset = in_bytes(cp_base_offset +
  2164                                     ConstantPoolCacheEntry::f2_offset());
  2166   if (is_invokevfinal) {
  2167     __ get_cache_and_index_at_bcp(Rcache, Rscratch, 1);
  2168     __ ld_ptr(Rcache, method_offset, Rmethod);
  2169   } else if (byte_no == f1_oop) {
  2170     // Resolved f1_oop goes directly into 'method' register.
  2171     resolve_cache_and_index(byte_no, Rmethod, Rcache, Rscratch, sizeof(u4));
  2172   } else {
  2173     resolve_cache_and_index(byte_no, noreg, Rcache, Rscratch, sizeof(u2));
  2174     __ ld_ptr(Rcache, method_offset, Rmethod);
  2177   if (Ritable_index != noreg) {
  2178     __ ld_ptr(Rcache, index_offset, Ritable_index);
  2180   __ ld_ptr(Rcache, flags_offset, Rflags);
  2183 // The Rcache register must be set before call
  2184 void TemplateTable::load_field_cp_cache_entry(Register Robj,
  2185                                               Register Rcache,
  2186                                               Register index,
  2187                                               Register Roffset,
  2188                                               Register Rflags,
  2189                                               bool is_static) {
  2190   assert_different_registers(Rcache, Rflags, Roffset);
  2192   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2194   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2195   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2196   if (is_static) {
  2197     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
  2201 // The registers Rcache and index expected to be set before call.
  2202 // Correct values of the Rcache and index registers are preserved.
  2203 void TemplateTable::jvmti_post_field_access(Register Rcache,
  2204                                             Register index,
  2205                                             bool is_static,
  2206                                             bool has_tos) {
  2207   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2209   if (JvmtiExport::can_post_field_access()) {
  2210     // Check to see if a field access watch has been set before we take
  2211     // the time to call into the VM.
  2212     Label Label1;
  2213     assert_different_registers(Rcache, index, G1_scratch);
  2214     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
  2215     __ load_contents(get_field_access_count_addr, G1_scratch);
  2216     __ tst(G1_scratch);
  2217     __ br(Assembler::zero, false, Assembler::pt, Label1);
  2218     __ delayed()->nop();
  2220     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
  2222     if (is_static) {
  2223       __ clr(Otos_i);
  2224     } else {
  2225       if (has_tos) {
  2226       // save object pointer before call_VM() clobbers it
  2227         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
  2228       } else {
  2229         // Load top of stack (do not pop the value off the stack);
  2230         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
  2232       __ verify_oop(Otos_i);
  2234     // Otos_i: object pointer or NULL if static
  2235     // Rcache: cache entry pointer
  2236     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
  2237                Otos_i, Rcache);
  2238     if (!is_static && has_tos) {
  2239       __ pop_ptr(Otos_i);  // restore object pointer
  2240       __ verify_oop(Otos_i);
  2242     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2243     __ bind(Label1);
  2247 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
  2248   transition(vtos, vtos);
  2250   Register Rcache = G3_scratch;
  2251   Register index  = G4_scratch;
  2252   Register Rclass = Rcache;
  2253   Register Roffset= G4_scratch;
  2254   Register Rflags = G1_scratch;
  2255   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2257   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
  2258   jvmti_post_field_access(Rcache, index, is_static, false);
  2259   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2261   if (!is_static) {
  2262     pop_and_check_object(Rclass);
  2263   } else {
  2264     __ verify_oop(Rclass);
  2267   Label exit;
  2269   Assembler::Membar_mask_bits membar_bits =
  2270     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2272   if (__ membar_has_effect(membar_bits)) {
  2273     // Get volatile flag
  2274     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2275     __ and3(Rflags, Lscratch, Lscratch);
  2278   Label checkVolatile;
  2280   // compute field type
  2281   Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
  2282   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2283   // Make sure we don't need to mask Rflags for tosBits after the above shift
  2284   ConstantPoolCacheEntry::verify_tosBits();
  2286   // Check atos before itos for getstatic, more likely (in Queens at least)
  2287   __ cmp(Rflags, atos);
  2288   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2289   __ delayed() ->cmp(Rflags, itos);
  2291   // atos
  2292   __ load_heap_oop(Rclass, Roffset, Otos_i);
  2293   __ verify_oop(Otos_i);
  2294   __ push(atos);
  2295   if (!is_static) {
  2296     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
  2298   __ ba(false, checkVolatile);
  2299   __ delayed()->tst(Lscratch);
  2301   __ bind(notObj);
  2303   // cmp(Rflags, itos);
  2304   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2305   __ delayed() ->cmp(Rflags, ltos);
  2307   // itos
  2308   __ ld(Rclass, Roffset, Otos_i);
  2309   __ push(itos);
  2310   if (!is_static) {
  2311     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
  2313   __ ba(false, checkVolatile);
  2314   __ delayed()->tst(Lscratch);
  2316   __ bind(notInt);
  2318   // cmp(Rflags, ltos);
  2319   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2320   __ delayed() ->cmp(Rflags, btos);
  2322   // ltos
  2323   // load must be atomic
  2324   __ ld_long(Rclass, Roffset, Otos_l);
  2325   __ push(ltos);
  2326   if (!is_static) {
  2327     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
  2329   __ ba(false, checkVolatile);
  2330   __ delayed()->tst(Lscratch);
  2332   __ bind(notLong);
  2334   // cmp(Rflags, btos);
  2335   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2336   __ delayed() ->cmp(Rflags, ctos);
  2338   // btos
  2339   __ ldsb(Rclass, Roffset, Otos_i);
  2340   __ push(itos);
  2341   if (!is_static) {
  2342     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
  2344   __ ba(false, checkVolatile);
  2345   __ delayed()->tst(Lscratch);
  2347   __ bind(notByte);
  2349   // cmp(Rflags, ctos);
  2350   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2351   __ delayed() ->cmp(Rflags, stos);
  2353   // ctos
  2354   __ lduh(Rclass, Roffset, Otos_i);
  2355   __ push(itos);
  2356   if (!is_static) {
  2357     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
  2359   __ ba(false, checkVolatile);
  2360   __ delayed()->tst(Lscratch);
  2362   __ bind(notChar);
  2364   // cmp(Rflags, stos);
  2365   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2366   __ delayed() ->cmp(Rflags, ftos);
  2368   // stos
  2369   __ ldsh(Rclass, Roffset, Otos_i);
  2370   __ push(itos);
  2371   if (!is_static) {
  2372     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
  2374   __ ba(false, checkVolatile);
  2375   __ delayed()->tst(Lscratch);
  2377   __ bind(notShort);
  2380   // cmp(Rflags, ftos);
  2381   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
  2382   __ delayed() ->tst(Lscratch);
  2384   // ftos
  2385   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
  2386   __ push(ftos);
  2387   if (!is_static) {
  2388     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
  2390   __ ba(false, checkVolatile);
  2391   __ delayed()->tst(Lscratch);
  2393   __ bind(notFloat);
  2396   // dtos
  2397   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
  2398   __ push(dtos);
  2399   if (!is_static) {
  2400     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
  2403   __ bind(checkVolatile);
  2404   if (__ membar_has_effect(membar_bits)) {
  2405     // __ tst(Lscratch); executed in delay slot
  2406     __ br(Assembler::zero, false, Assembler::pt, exit);
  2407     __ delayed()->nop();
  2408     volatile_barrier(membar_bits);
  2411   __ bind(exit);
  2415 void TemplateTable::getfield(int byte_no) {
  2416   getfield_or_static(byte_no, false);
  2419 void TemplateTable::getstatic(int byte_no) {
  2420   getfield_or_static(byte_no, true);
  2424 void TemplateTable::fast_accessfield(TosState state) {
  2425   transition(atos, state);
  2426   Register Rcache  = G3_scratch;
  2427   Register index   = G4_scratch;
  2428   Register Roffset = G4_scratch;
  2429   Register Rflags  = Rcache;
  2430   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2432   __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2433   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
  2435   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2437   __ null_check(Otos_i);
  2438   __ verify_oop(Otos_i);
  2440   Label exit;
  2442   Assembler::Membar_mask_bits membar_bits =
  2443     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2444   if (__ membar_has_effect(membar_bits)) {
  2445     // Get volatile flag
  2446     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
  2447     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2450   switch (bytecode()) {
  2451     case Bytecodes::_fast_bgetfield:
  2452       __ ldsb(Otos_i, Roffset, Otos_i);
  2453       break;
  2454     case Bytecodes::_fast_cgetfield:
  2455       __ lduh(Otos_i, Roffset, Otos_i);
  2456       break;
  2457     case Bytecodes::_fast_sgetfield:
  2458       __ ldsh(Otos_i, Roffset, Otos_i);
  2459       break;
  2460     case Bytecodes::_fast_igetfield:
  2461       __ ld(Otos_i, Roffset, Otos_i);
  2462       break;
  2463     case Bytecodes::_fast_lgetfield:
  2464       __ ld_long(Otos_i, Roffset, Otos_l);
  2465       break;
  2466     case Bytecodes::_fast_fgetfield:
  2467       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
  2468       break;
  2469     case Bytecodes::_fast_dgetfield:
  2470       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
  2471       break;
  2472     case Bytecodes::_fast_agetfield:
  2473       __ load_heap_oop(Otos_i, Roffset, Otos_i);
  2474       break;
  2475     default:
  2476       ShouldNotReachHere();
  2479   if (__ membar_has_effect(membar_bits)) {
  2480     __ btst(Lscratch, Rflags);
  2481     __ br(Assembler::zero, false, Assembler::pt, exit);
  2482     __ delayed()->nop();
  2483     volatile_barrier(membar_bits);
  2484     __ bind(exit);
  2487   if (state == atos) {
  2488     __ verify_oop(Otos_i);    // does not blow flags!
  2492 void TemplateTable::jvmti_post_fast_field_mod() {
  2493   if (JvmtiExport::can_post_field_modification()) {
  2494     // Check to see if a field modification watch has been set before we take
  2495     // the time to call into the VM.
  2496     Label done;
  2497     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2498     __ load_contents(get_field_modification_count_addr, G4_scratch);
  2499     __ tst(G4_scratch);
  2500     __ br(Assembler::zero, false, Assembler::pt, done);
  2501     __ delayed()->nop();
  2502     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
  2503     __ verify_oop(G4_scratch);
  2504     __ push_ptr(G4_scratch);    // put the object pointer back on tos
  2505     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
  2506     // Save tos values before call_VM() clobbers them. Since we have
  2507     // to do it for every data type, we use the saved values as the
  2508     // jvalue object.
  2509     switch (bytecode()) {  // save tos values before call_VM() clobbers them
  2510     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
  2511     case Bytecodes::_fast_bputfield: // fall through
  2512     case Bytecodes::_fast_sputfield: // fall through
  2513     case Bytecodes::_fast_cputfield: // fall through
  2514     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
  2515     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
  2516     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
  2517     // get words in right order for use as jvalue object
  2518     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
  2520     // setup pointer to jvalue object
  2521     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
  2522     // G4_scratch:  object pointer
  2523     // G1_scratch: cache entry pointer
  2524     // G3_scratch: jvalue object on the stack
  2525     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
  2526     switch (bytecode()) {             // restore tos values
  2527     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
  2528     case Bytecodes::_fast_bputfield: // fall through
  2529     case Bytecodes::_fast_sputfield: // fall through
  2530     case Bytecodes::_fast_cputfield: // fall through
  2531     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
  2532     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
  2533     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
  2534     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
  2536     __ bind(done);
  2540 // The registers Rcache and index expected to be set before call.
  2541 // The function may destroy various registers, just not the Rcache and index registers.
  2542 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
  2543   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2545   if (JvmtiExport::can_post_field_modification()) {
  2546     // Check to see if a field modification watch has been set before we take
  2547     // the time to call into the VM.
  2548     Label Label1;
  2549     assert_different_registers(Rcache, index, G1_scratch);
  2550     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2551     __ load_contents(get_field_modification_count_addr, G1_scratch);
  2552     __ tst(G1_scratch);
  2553     __ br(Assembler::zero, false, Assembler::pt, Label1);
  2554     __ delayed()->nop();
  2556     // The Rcache and index registers have been already set.
  2557     // This allows to eliminate this call but the Rcache and index
  2558     // registers must be correspondingly used after this line.
  2559     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
  2561     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
  2562     if (is_static) {
  2563       // Life is simple.  Null out the object pointer.
  2564       __ clr(G4_scratch);
  2565     } else {
  2566       Register Rflags = G1_scratch;
  2567       // Life is harder. The stack holds the value on top, followed by the
  2568       // object.  We don't know the size of the value, though; it could be
  2569       // one or two words depending on its type. As a result, we must find
  2570       // the type to determine where the object is.
  2572       Label two_word, valsizeknown;
  2573       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2574       __ mov(Lesp, G4_scratch);
  2575       __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2576       // Make sure we don't need to mask Rflags for tosBits after the above shift
  2577       ConstantPoolCacheEntry::verify_tosBits();
  2578       __ cmp(Rflags, ltos);
  2579       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2580       __ delayed()->cmp(Rflags, dtos);
  2581       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2582       __ delayed()->nop();
  2583       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
  2584       __ br(Assembler::always, false, Assembler::pt, valsizeknown);
  2585       __ delayed()->nop();
  2586       __ bind(two_word);
  2588       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
  2590       __ bind(valsizeknown);
  2591       // setup object pointer
  2592       __ ld_ptr(G4_scratch, 0, G4_scratch);
  2593       __ verify_oop(G4_scratch);
  2595     // setup pointer to jvalue object
  2596     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
  2597     // G4_scratch:  object pointer or NULL if static
  2598     // G3_scratch: cache entry pointer
  2599     // G1_scratch: jvalue object on the stack
  2600     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
  2601                G4_scratch, G3_scratch, G1_scratch);
  2602     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2603     __ bind(Label1);
  2607 void TemplateTable::pop_and_check_object(Register r) {
  2608   __ pop_ptr(r);
  2609   __ null_check(r);  // for field access must check obj.
  2610   __ verify_oop(r);
  2613 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
  2614   transition(vtos, vtos);
  2615   Register Rcache = G3_scratch;
  2616   Register index  = G4_scratch;
  2617   Register Rclass = Rcache;
  2618   Register Roffset= G4_scratch;
  2619   Register Rflags = G1_scratch;
  2620   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2622   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
  2623   jvmti_post_field_mod(Rcache, index, is_static);
  2624   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2626   Assembler::Membar_mask_bits read_bits =
  2627     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2628   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2630   Label notVolatile, checkVolatile, exit;
  2631   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2632     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2633     __ and3(Rflags, Lscratch, Lscratch);
  2635     if (__ membar_has_effect(read_bits)) {
  2636       __ tst(Lscratch);
  2637       __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2638       __ delayed()->nop();
  2639       volatile_barrier(read_bits);
  2640       __ bind(notVolatile);
  2644   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2645   // Make sure we don't need to mask Rflags for tosBits after the above shift
  2646   ConstantPoolCacheEntry::verify_tosBits();
  2648   // compute field type
  2649   Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
  2651   if (is_static) {
  2652     // putstatic with object type most likely, check that first
  2653     __ cmp(Rflags, atos );
  2654     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2655     __ delayed() ->cmp(Rflags, itos );
  2657     // atos
  2658     __ pop_ptr();
  2659     __ verify_oop(Otos_i);
  2661     do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2663     __ ba(false, checkVolatile);
  2664     __ delayed()->tst(Lscratch);
  2666     __ bind(notObj);
  2668     // cmp(Rflags, itos );
  2669     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2670     __ delayed() ->cmp(Rflags, btos );
  2672     // itos
  2673     __ pop_i();
  2674     __ st(Otos_i, Rclass, Roffset);
  2675     __ ba(false, checkVolatile);
  2676     __ delayed()->tst(Lscratch);
  2678     __ bind(notInt);
  2680   } else {
  2681     // putfield with int type most likely, check that first
  2682     __ cmp(Rflags, itos );
  2683     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2684     __ delayed() ->cmp(Rflags, atos );
  2686     // itos
  2687     __ pop_i();
  2688     pop_and_check_object(Rclass);
  2689     __ st(Otos_i, Rclass, Roffset);
  2690     patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch);
  2691     __ ba(false, checkVolatile);
  2692     __ delayed()->tst(Lscratch);
  2694     __ bind(notInt);
  2695     // cmp(Rflags, atos );
  2696     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2697     __ delayed() ->cmp(Rflags, btos );
  2699     // atos
  2700     __ pop_ptr();
  2701     pop_and_check_object(Rclass);
  2702     __ verify_oop(Otos_i);
  2704     do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2706     patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch);
  2707     __ ba(false, checkVolatile);
  2708     __ delayed()->tst(Lscratch);
  2710     __ bind(notObj);
  2713   // cmp(Rflags, btos );
  2714   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2715   __ delayed() ->cmp(Rflags, ltos );
  2717   // btos
  2718   __ pop_i();
  2719   if (!is_static) pop_and_check_object(Rclass);
  2720   __ stb(Otos_i, Rclass, Roffset);
  2721   if (!is_static) {
  2722     patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch);
  2724   __ ba(false, checkVolatile);
  2725   __ delayed()->tst(Lscratch);
  2727   __ bind(notByte);
  2729   // cmp(Rflags, ltos );
  2730   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2731   __ delayed() ->cmp(Rflags, ctos );
  2733   // ltos
  2734   __ pop_l();
  2735   if (!is_static) pop_and_check_object(Rclass);
  2736   __ st_long(Otos_l, Rclass, Roffset);
  2737   if (!is_static) {
  2738     patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch);
  2740   __ ba(false, checkVolatile);
  2741   __ delayed()->tst(Lscratch);
  2743   __ bind(notLong);
  2745   // cmp(Rflags, ctos );
  2746   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2747   __ delayed() ->cmp(Rflags, stos );
  2749   // ctos (char)
  2750   __ pop_i();
  2751   if (!is_static) pop_and_check_object(Rclass);
  2752   __ sth(Otos_i, Rclass, Roffset);
  2753   if (!is_static) {
  2754     patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch);
  2756   __ ba(false, checkVolatile);
  2757   __ delayed()->tst(Lscratch);
  2759   __ bind(notChar);
  2760   // cmp(Rflags, stos );
  2761   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2762   __ delayed() ->cmp(Rflags, ftos );
  2764   // stos (char)
  2765   __ pop_i();
  2766   if (!is_static) pop_and_check_object(Rclass);
  2767   __ sth(Otos_i, Rclass, Roffset);
  2768   if (!is_static) {
  2769     patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch);
  2771   __ ba(false, checkVolatile);
  2772   __ delayed()->tst(Lscratch);
  2774   __ bind(notShort);
  2775   // cmp(Rflags, ftos );
  2776   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
  2777   __ delayed()->nop();
  2779   // ftos
  2780   __ pop_f();
  2781   if (!is_static) pop_and_check_object(Rclass);
  2782   __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2783   if (!is_static) {
  2784     patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch);
  2786   __ ba(false, checkVolatile);
  2787   __ delayed()->tst(Lscratch);
  2789   __ bind(notFloat);
  2791   // dtos
  2792   __ pop_d();
  2793   if (!is_static) pop_and_check_object(Rclass);
  2794   __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2795   if (!is_static) {
  2796     patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch);
  2799   __ bind(checkVolatile);
  2800   __ tst(Lscratch);
  2802   if (__ membar_has_effect(write_bits)) {
  2803     // __ tst(Lscratch); in delay slot
  2804     __ br(Assembler::zero, false, Assembler::pt, exit);
  2805     __ delayed()->nop();
  2806     volatile_barrier(Assembler::StoreLoad);
  2807     __ bind(exit);
  2811 void TemplateTable::fast_storefield(TosState state) {
  2812   transition(state, vtos);
  2813   Register Rcache = G3_scratch;
  2814   Register Rclass = Rcache;
  2815   Register Roffset= G4_scratch;
  2816   Register Rflags = G1_scratch;
  2817   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2819   jvmti_post_fast_field_mod();
  2821   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
  2823   Assembler::Membar_mask_bits read_bits =
  2824     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2825   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2827   Label notVolatile, checkVolatile, exit;
  2828   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2829     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2830     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2831     __ and3(Rflags, Lscratch, Lscratch);
  2832     if (__ membar_has_effect(read_bits)) {
  2833       __ tst(Lscratch);
  2834       __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2835       __ delayed()->nop();
  2836       volatile_barrier(read_bits);
  2837       __ bind(notVolatile);
  2841   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2842   pop_and_check_object(Rclass);
  2844   switch (bytecode()) {
  2845     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
  2846     case Bytecodes::_fast_cputfield: /* fall through */
  2847     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
  2848     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
  2849     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
  2850     case Bytecodes::_fast_fputfield:
  2851       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2852       break;
  2853     case Bytecodes::_fast_dputfield:
  2854       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2855       break;
  2856     case Bytecodes::_fast_aputfield:
  2857       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2858       break;
  2859     default:
  2860       ShouldNotReachHere();
  2863   if (__ membar_has_effect(write_bits)) {
  2864     __ tst(Lscratch);
  2865     __ br(Assembler::zero, false, Assembler::pt, exit);
  2866     __ delayed()->nop();
  2867     volatile_barrier(Assembler::StoreLoad);
  2868     __ bind(exit);
  2873 void TemplateTable::putfield(int byte_no) {
  2874   putfield_or_static(byte_no, false);
  2877 void TemplateTable::putstatic(int byte_no) {
  2878   putfield_or_static(byte_no, true);
  2882 void TemplateTable::fast_xaccess(TosState state) {
  2883   transition(vtos, state);
  2884   Register Rcache = G3_scratch;
  2885   Register Roffset = G4_scratch;
  2886   Register Rflags  = G4_scratch;
  2887   Register Rreceiver = Lscratch;
  2889   __ ld_ptr(Llocals, 0, Rreceiver);
  2891   // access constant pool cache  (is resolved)
  2892   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
  2893   __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2894   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
  2896   __ verify_oop(Rreceiver);
  2897   __ null_check(Rreceiver);
  2898   if (state == atos) {
  2899     __ load_heap_oop(Rreceiver, Roffset, Otos_i);
  2900   } else if (state == itos) {
  2901     __ ld (Rreceiver, Roffset, Otos_i) ;
  2902   } else if (state == ftos) {
  2903     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
  2904   } else {
  2905     ShouldNotReachHere();
  2908   Assembler::Membar_mask_bits membar_bits =
  2909     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2910   if (__ membar_has_effect(membar_bits)) {
  2912     // Get is_volatile value in Rflags and check if membar is needed
  2913     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2915     // Test volatile
  2916     Label notVolatile;
  2917     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2918     __ btst(Rflags, Lscratch);
  2919     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2920     __ delayed()->nop();
  2921     volatile_barrier(membar_bits);
  2922     __ bind(notVolatile);
  2925   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
  2926   __ sub(Lbcp, 1, Lbcp);
  2929 //----------------------------------------------------------------------------------------------------
  2930 // Calls
  2932 void TemplateTable::count_calls(Register method, Register temp) {
  2933   // implemented elsewhere
  2934   ShouldNotReachHere();
  2937 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
  2938   Register Rtemp = G4_scratch;
  2939   Register Rcall = Rindex;
  2940   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  2942   // get target methodOop & entry point
  2943   const int base = instanceKlass::vtable_start_offset() * wordSize;
  2944   if (vtableEntry::size() % 3 == 0) {
  2945     // scale the vtable index by 12:
  2946     int one_third = vtableEntry::size() / 3;
  2947     __ sll(Rindex, exact_log2(one_third * 1 * wordSize), Rtemp);
  2948     __ sll(Rindex, exact_log2(one_third * 2 * wordSize), Rindex);
  2949     __ add(Rindex, Rtemp, Rindex);
  2950   } else {
  2951     // scale the vtable index by 8:
  2952     __ sll(Rindex, exact_log2(vtableEntry::size() * wordSize), Rindex);
  2955   __ add(Rrecv, Rindex, Rrecv);
  2956   __ ld_ptr(Rrecv, base + vtableEntry::method_offset_in_bytes(), G5_method);
  2958   __ call_from_interpreter(Rcall, Gargs, Rret);
  2961 void TemplateTable::invokevirtual(int byte_no) {
  2962   transition(vtos, vtos);
  2963   assert(byte_no == f2_byte, "use this argument");
  2965   Register Rscratch = G3_scratch;
  2966   Register Rtemp = G4_scratch;
  2967   Register Rret = Lscratch;
  2968   Register Rrecv = G5_method;
  2969   Label notFinal;
  2971   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
  2972   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  2974   // Check for vfinal
  2975   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), G4_scratch);
  2976   __ btst(Rret, G4_scratch);
  2977   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  2978   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
  2980   patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
  2982   invokevfinal_helper(Rscratch, Rret);
  2984   __ bind(notFinal);
  2986   __ mov(G5_method, Rscratch);  // better scratch register
  2987   __ load_receiver(G4_scratch, O0);  // gets receiverOop
  2988   // receiver is in O0
  2989   __ verify_oop(O0);
  2991   // get return address
  2992   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  2993   __ set(table, Rtemp);
  2994   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  2995   // Make sure we don't need to mask Rret for tosBits after the above shift
  2996   ConstantPoolCacheEntry::verify_tosBits();
  2997   __ sll(Rret,  LogBytesPerWord, Rret);
  2998   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3000   // get receiver klass
  3001   __ null_check(O0, oopDesc::klass_offset_in_bytes());
  3002   __ load_klass(O0, Rrecv);
  3003   __ verify_oop(Rrecv);
  3005   __ profile_virtual_call(Rrecv, O4);
  3007   generate_vtable_call(Rrecv, Rscratch, Rret);
  3010 void TemplateTable::fast_invokevfinal(int byte_no) {
  3011   transition(vtos, vtos);
  3012   assert(byte_no == f2_byte, "use this argument");
  3014   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
  3015                              /*is_invokevfinal*/true, false);
  3016   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3017   invokevfinal_helper(G3_scratch, Lscratch);
  3020 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
  3021   Register Rtemp = G4_scratch;
  3023   __ verify_oop(G5_method);
  3025   // Load receiver from stack slot
  3026   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
  3027   __ load_receiver(G4_scratch, O0);
  3029   // receiver NULL check
  3030   __ null_check(O0);
  3032   __ profile_final_call(O4);
  3034   // get return address
  3035   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3036   __ set(table, Rtemp);
  3037   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3038   // Make sure we don't need to mask Rret for tosBits after the above shift
  3039   ConstantPoolCacheEntry::verify_tosBits();
  3040   __ sll(Rret,  LogBytesPerWord, Rret);
  3041   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3044   // do the call
  3045   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3048 void TemplateTable::invokespecial(int byte_no) {
  3049   transition(vtos, vtos);
  3050   assert(byte_no == f1_byte, "use this argument");
  3052   Register Rscratch = G3_scratch;
  3053   Register Rtemp = G4_scratch;
  3054   Register Rret = Lscratch;
  3056   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
  3057   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3059   __ verify_oop(G5_method);
  3061   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
  3062   __ load_receiver(G4_scratch, O0);
  3064   // receiver NULL check
  3065   __ null_check(O0);
  3067   __ profile_call(O4);
  3069   // get return address
  3070   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3071   __ set(table, Rtemp);
  3072   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3073   // Make sure we don't need to mask Rret for tosBits after the above shift
  3074   ConstantPoolCacheEntry::verify_tosBits();
  3075   __ sll(Rret,  LogBytesPerWord, Rret);
  3076   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3078   // do the call
  3079   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3082 void TemplateTable::invokestatic(int byte_no) {
  3083   transition(vtos, vtos);
  3084   assert(byte_no == f1_byte, "use this argument");
  3086   Register Rscratch = G3_scratch;
  3087   Register Rtemp = G4_scratch;
  3088   Register Rret = Lscratch;
  3090   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
  3091   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3093   __ verify_oop(G5_method);
  3095   __ profile_call(O4);
  3097   // get return address
  3098   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3099   __ set(table, Rtemp);
  3100   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3101   // Make sure we don't need to mask Rret for tosBits after the above shift
  3102   ConstantPoolCacheEntry::verify_tosBits();
  3103   __ sll(Rret,  LogBytesPerWord, Rret);
  3104   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3106   // do the call
  3107   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3111 void TemplateTable::invokeinterface_object_method(Register RklassOop,
  3112                                                   Register Rcall,
  3113                                                   Register Rret,
  3114                                                   Register Rflags) {
  3115   Register Rscratch = G4_scratch;
  3116   Register Rindex = Lscratch;
  3118   assert_different_registers(Rscratch, Rindex, Rret);
  3120   Label notFinal;
  3122   // Check for vfinal
  3123   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), Rscratch);
  3124   __ btst(Rflags, Rscratch);
  3125   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  3126   __ delayed()->nop();
  3128   __ profile_final_call(O4);
  3130   // do the call - the index (f2) contains the methodOop
  3131   assert_different_registers(G5_method, Gargs, Rcall);
  3132   __ mov(Rindex, G5_method);
  3133   __ call_from_interpreter(Rcall, Gargs, Rret);
  3134   __ bind(notFinal);
  3136   __ profile_virtual_call(RklassOop, O4);
  3137   generate_vtable_call(RklassOop, Rindex, Rret);
  3141 void TemplateTable::invokeinterface(int byte_no) {
  3142   transition(vtos, vtos);
  3143   assert(byte_no == f1_byte, "use this argument");
  3145   Register Rscratch = G4_scratch;
  3146   Register Rret = G3_scratch;
  3147   Register Rindex = Lscratch;
  3148   Register Rinterface = G1_scratch;
  3149   Register RklassOop = G5_method;
  3150   Register Rflags = O1;
  3151   assert_different_registers(Rscratch, G5_method);
  3153   load_invoke_cp_cache_entry(byte_no, Rinterface, Rindex, Rflags, /*virtual*/ false, false, false);
  3154   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3156   // get receiver
  3157   __ and3(Rflags, 0xFF, Rscratch);       // gets number of parameters
  3158   __ load_receiver(Rscratch, O0);
  3159   __ verify_oop(O0);
  3161   __ mov(Rflags, Rret);
  3163   // get return address
  3164   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
  3165   __ set(table, Rscratch);
  3166   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3167   // Make sure we don't need to mask Rret for tosBits after the above shift
  3168   ConstantPoolCacheEntry::verify_tosBits();
  3169   __ sll(Rret,  LogBytesPerWord, Rret);
  3170   __ ld_ptr(Rscratch, Rret, Rret);      // get return address
  3172   // get receiver klass
  3173   __ null_check(O0, oopDesc::klass_offset_in_bytes());
  3174   __ load_klass(O0, RklassOop);
  3175   __ verify_oop(RklassOop);
  3177   // Special case of invokeinterface called for virtual method of
  3178   // java.lang.Object.  See cpCacheOop.cpp for details.
  3179   // This code isn't produced by javac, but could be produced by
  3180   // another compliant java compiler.
  3181   Label notMethod;
  3182   __ set((1 << ConstantPoolCacheEntry::methodInterface), Rscratch);
  3183   __ btst(Rflags, Rscratch);
  3184   __ br(Assembler::zero, false, Assembler::pt, notMethod);
  3185   __ delayed()->nop();
  3187   invokeinterface_object_method(RklassOop, Rinterface, Rret, Rflags);
  3189   __ bind(notMethod);
  3191   __ profile_virtual_call(RklassOop, O4);
  3193   //
  3194   // find entry point to call
  3195   //
  3197   // compute start of first itableOffsetEntry (which is at end of vtable)
  3198   const int base = instanceKlass::vtable_start_offset() * wordSize;
  3199   Label search;
  3200   Register Rtemp = Rflags;
  3202   __ ld(RklassOop, instanceKlass::vtable_length_offset() * wordSize, Rtemp);
  3203   if (align_object_offset(1) > 1) {
  3204     __ round_to(Rtemp, align_object_offset(1));
  3206   __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
  3207   if (Assembler::is_simm13(base)) {
  3208     __ add(Rtemp, base, Rtemp);
  3209   } else {
  3210     __ set(base, Rscratch);
  3211     __ add(Rscratch, Rtemp, Rtemp);
  3213   __ add(RklassOop, Rtemp, Rscratch);
  3215   __ bind(search);
  3217   __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
  3219     Label ok;
  3221     // Check that entry is non-null.  Null entries are probably a bytecode
  3222     // problem.  If the interface isn't implemented by the receiver class,
  3223     // the VM should throw IncompatibleClassChangeError.  linkResolver checks
  3224     // this too but that's only if the entry isn't already resolved, so we
  3225     // need to check again.
  3226     __ br_notnull( Rtemp, false, Assembler::pt, ok);
  3227     __ delayed()->nop();
  3228     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
  3229     __ should_not_reach_here();
  3230     __ bind(ok);
  3231     __ verify_oop(Rtemp);
  3234   __ verify_oop(Rinterface);
  3236   __ cmp(Rinterface, Rtemp);
  3237   __ brx(Assembler::notEqual, true, Assembler::pn, search);
  3238   __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
  3240   // entry found and Rscratch points to it
  3241   __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
  3243   assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
  3244   __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
  3245   __ add(Rscratch, Rindex, Rscratch);
  3246   __ ld_ptr(RklassOop, Rscratch, G5_method);
  3248   // Check for abstract method error.
  3250     Label ok;
  3251     __ tst(G5_method);
  3252     __ brx(Assembler::notZero, false, Assembler::pt, ok);
  3253     __ delayed()->nop();
  3254     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
  3255     __ should_not_reach_here();
  3256     __ bind(ok);
  3259   Register Rcall = Rinterface;
  3260   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  3262   __ verify_oop(G5_method);
  3263   __ call_from_interpreter(Rcall, Gargs, Rret);
  3268 void TemplateTable::invokedynamic(int byte_no) {
  3269   transition(vtos, vtos);
  3270   assert(byte_no == f1_oop, "use this argument");
  3272   if (!EnableInvokeDynamic) {
  3273     // We should not encounter this bytecode if !EnableInvokeDynamic.
  3274     // The verifier will stop it.  However, if we get past the verifier,
  3275     // this will stop the thread in a reasonable way, without crashing the JVM.
  3276     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
  3277                      InterpreterRuntime::throw_IncompatibleClassChangeError));
  3278     // the call_VM checks for exception, so we should never return here.
  3279     __ should_not_reach_here();
  3280     return;
  3283   // G5: CallSite object (f1)
  3284   // XX: unused (f2)
  3285   // XX: flags (unused)
  3287   Register G5_callsite = G5_method;
  3288   Register Rscratch    = G3_scratch;
  3289   Register Rtemp       = G1_scratch;
  3290   Register Rret        = Lscratch;
  3292   load_invoke_cp_cache_entry(byte_no, G5_callsite, noreg, Rret,
  3293                              /*virtual*/ false, /*vfinal*/ false, /*indy*/ true);
  3294   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
  3296   // profile this call
  3297   __ profile_call(O4);
  3299   // get return address
  3300   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
  3301   __ set(table, Rtemp);
  3302   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);  // get return type
  3303   // Make sure we don't need to mask Rret for tosBits after the above shift
  3304   ConstantPoolCacheEntry::verify_tosBits();
  3305   __ sll(Rret, LogBytesPerWord, Rret);
  3306   __ ld_ptr(Rtemp, Rret, Rret);  // get return address
  3308   __ verify_oop(G5_callsite);
  3309   __ load_heap_oop(G5_callsite, __ delayed_value(java_lang_invoke_CallSite::target_offset_in_bytes, Rscratch), G3_method_handle);
  3310   __ null_check(G3_method_handle);
  3311   __ verify_oop(G3_method_handle);
  3313   // Adjust Rret first so Llast_SP can be same as Rret
  3314   __ add(Rret, -frame::pc_return_offset, O7);
  3315   __ add(Lesp, BytesPerWord, Gargs);  // setup parameter pointer
  3316   __ jump_to_method_handle_entry(G3_method_handle, Rtemp, /* emit_delayed_nop */ false);
  3317   // Record SP so we can remove any stack space allocated by adapter transition
  3318   __ delayed()->mov(SP, Llast_SP);
  3322 //----------------------------------------------------------------------------------------------------
  3323 // Allocation
  3325 void TemplateTable::_new() {
  3326   transition(vtos, atos);
  3328   Label slow_case;
  3329   Label done;
  3330   Label initialize_header;
  3331   Label initialize_object;  // including clearing the fields
  3333   Register RallocatedObject = Otos_i;
  3334   Register RinstanceKlass = O1;
  3335   Register Roffset = O3;
  3336   Register Rscratch = O4;
  3338   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3339   __ get_cpool_and_tags(Rscratch, G3_scratch);
  3340   // make sure the class we're about to instantiate has been resolved
  3341   // This is done before loading instanceKlass to be consistent with the order
  3342   // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
  3343   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3344   __ ldub(G3_scratch, Roffset, G3_scratch);
  3345   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3346   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3347   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3348   // get instanceKlass
  3349   //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
  3350   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3351   __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
  3353   // make sure klass is fully initialized:
  3354   __ ld(RinstanceKlass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_scratch);
  3355   __ cmp(G3_scratch, instanceKlass::fully_initialized);
  3356   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3357   __ delayed()->ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), Roffset);
  3359   // get instance_size in instanceKlass (already aligned)
  3360   //__ ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), Roffset);
  3362   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
  3363   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
  3364   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
  3365   __ delayed()->nop();
  3367   // allocate the instance
  3368   // 1) Try to allocate in the TLAB
  3369   // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
  3370   // 3) if the above fails (or is not applicable), go to a slow case
  3371   // (creates a new TLAB, etc.)
  3373   const bool allow_shared_alloc =
  3374     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
  3376   if(UseTLAB) {
  3377     Register RoldTopValue = RallocatedObject;
  3378     Register RtopAddr = G3_scratch, RtlabWasteLimitValue = G3_scratch;
  3379     Register RnewTopValue = G1_scratch;
  3380     Register RendValue = Rscratch;
  3381     Register RfreeValue = RnewTopValue;
  3383     // check if we can allocate in the TLAB
  3384     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
  3385     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
  3386     __ add(RoldTopValue, Roffset, RnewTopValue);
  3388     // if there is enough space, we do not CAS and do not clear
  3389     __ cmp(RnewTopValue, RendValue);
  3390     if(ZeroTLAB) {
  3391       // the fields have already been cleared
  3392       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
  3393     } else {
  3394       // initialize both the header and fields
  3395       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
  3397     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
  3399     if (allow_shared_alloc) {
  3400       // Check if tlab should be discarded (refill_waste_limit >= free)
  3401       __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
  3402       __ sub(RendValue, RoldTopValue, RfreeValue);
  3403 #ifdef _LP64
  3404       __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
  3405 #else
  3406       __ srl(RfreeValue, LogHeapWordSize, RfreeValue);
  3407 #endif
  3408       __ cmp(RtlabWasteLimitValue, RfreeValue);
  3409       __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, slow_case); // tlab waste is small
  3410       __ delayed()->nop();
  3412       // increment waste limit to prevent getting stuck on this slow path
  3413       __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
  3414       __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
  3415     } else {
  3416       // No allocation in the shared eden.
  3417       __ br(Assembler::always, false, Assembler::pt, slow_case);
  3418       __ delayed()->nop();
  3422   // Allocation in the shared Eden
  3423   if (allow_shared_alloc) {
  3424     Register RoldTopValue = G1_scratch;
  3425     Register RtopAddr = G3_scratch;
  3426     Register RnewTopValue = RallocatedObject;
  3427     Register RendValue = Rscratch;
  3429     __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
  3431     Label retry;
  3432     __ bind(retry);
  3433     __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
  3434     __ ld_ptr(RendValue, 0, RendValue);
  3435     __ ld_ptr(RtopAddr, 0, RoldTopValue);
  3436     __ add(RoldTopValue, Roffset, RnewTopValue);
  3438     // RnewTopValue contains the top address after the new object
  3439     // has been allocated.
  3440     __ cmp(RnewTopValue, RendValue);
  3441     __ brx(Assembler::greaterUnsigned, false, Assembler::pn, slow_case);
  3442     __ delayed()->nop();
  3444     __ casx_under_lock(RtopAddr, RoldTopValue, RnewTopValue,
  3445       VM_Version::v9_instructions_work() ? NULL :
  3446       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
  3448     // if someone beat us on the allocation, try again, otherwise continue
  3449     __ cmp(RoldTopValue, RnewTopValue);
  3450     __ brx(Assembler::notEqual, false, Assembler::pn, retry);
  3451     __ delayed()->nop();
  3453     // bump total bytes allocated by this thread
  3454     // RoldTopValue and RtopAddr are dead, so can use G1 and G3
  3455     __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
  3458   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
  3459     // clear object fields
  3460     __ bind(initialize_object);
  3461     __ deccc(Roffset, sizeof(oopDesc));
  3462     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
  3463     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
  3465     // initialize remaining object fields
  3466     { Label loop;
  3467       __ subcc(Roffset, wordSize, Roffset);
  3468       __ bind(loop);
  3469       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
  3470       __ st_ptr(G0, G3_scratch, Roffset);
  3471       __ br(Assembler::notEqual, false, Assembler::pt, loop);
  3472       __ delayed()->subcc(Roffset, wordSize, Roffset);
  3474     __ br(Assembler::always, false, Assembler::pt, initialize_header);
  3475     __ delayed()->nop();
  3478   // slow case
  3479   __ bind(slow_case);
  3480   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3481   __ get_constant_pool(O1);
  3483   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
  3485   __ ba(false, done);
  3486   __ delayed()->nop();
  3488   // Initialize the header: mark, klass
  3489   __ bind(initialize_header);
  3491   if (UseBiasedLocking) {
  3492     __ ld_ptr(RinstanceKlass, Klass::prototype_header_offset_in_bytes() + sizeof(oopDesc), G4_scratch);
  3493   } else {
  3494     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
  3496   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
  3497   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
  3498   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
  3501     SkipIfEqual skip_if(
  3502       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
  3503     // Trigger dtrace event
  3504     __ push(atos);
  3505     __ call_VM_leaf(noreg,
  3506        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
  3507     __ pop(atos);
  3510   // continue
  3511   __ bind(done);
  3516 void TemplateTable::newarray() {
  3517   transition(itos, atos);
  3518   __ ldub(Lbcp, 1, O1);
  3519      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
  3523 void TemplateTable::anewarray() {
  3524   transition(itos, atos);
  3525   __ get_constant_pool(O1);
  3526   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3527      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
  3531 void TemplateTable::arraylength() {
  3532   transition(atos, itos);
  3533   Label ok;
  3534   __ verify_oop(Otos_i);
  3535   __ tst(Otos_i);
  3536   __ throw_if_not_1_x( Assembler::notZero, ok );
  3537   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
  3538   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3542 void TemplateTable::checkcast() {
  3543   transition(atos, atos);
  3544   Label done, is_null, quicked, cast_ok, resolved;
  3545   Register Roffset = G1_scratch;
  3546   Register RobjKlass = O5;
  3547   Register RspecifiedKlass = O4;
  3549   // Check for casting a NULL
  3550   __ br_null(Otos_i, false, Assembler::pn, is_null);
  3551   __ delayed()->nop();
  3553   // Get value klass in RobjKlass
  3554   __ load_klass(Otos_i, RobjKlass); // get value klass
  3556   // Get constant pool tag
  3557   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3559   // See if the checkcast has been quickened
  3560   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3561   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3562   __ ldub(G3_scratch, Roffset, G3_scratch);
  3563   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3564   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3565   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3567   __ push_ptr(); // save receiver for result, and for GC
  3568   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3569   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3571   __ br(Assembler::always, false, Assembler::pt, resolved);
  3572   __ delayed()->nop();
  3574   // Extract target class from constant pool
  3575   __ bind(quicked);
  3576   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3577   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3578   __ bind(resolved);
  3579   __ load_klass(Otos_i, RobjKlass); // get value klass
  3581   // Generate a fast subtype check.  Branch to cast_ok if no
  3582   // failure.  Throw exception if failure.
  3583   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
  3585   // Not a subtype; so must throw exception
  3586   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
  3588   __ bind(cast_ok);
  3590   if (ProfileInterpreter) {
  3591     __ ba(false, done);
  3592     __ delayed()->nop();
  3594   __ bind(is_null);
  3595   __ profile_null_seen(G3_scratch);
  3596   __ bind(done);
  3600 void TemplateTable::instanceof() {
  3601   Label done, is_null, quicked, resolved;
  3602   transition(atos, itos);
  3603   Register Roffset = G1_scratch;
  3604   Register RobjKlass = O5;
  3605   Register RspecifiedKlass = O4;
  3607   // Check for casting a NULL
  3608   __ br_null(Otos_i, false, Assembler::pt, is_null);
  3609   __ delayed()->nop();
  3611   // Get value klass in RobjKlass
  3612   __ load_klass(Otos_i, RobjKlass); // get value klass
  3614   // Get constant pool tag
  3615   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3617   // See if the checkcast has been quickened
  3618   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3619   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3620   __ ldub(G3_scratch, Roffset, G3_scratch);
  3621   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3622   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3623   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3625   __ push_ptr(); // save receiver for result, and for GC
  3626   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3627   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3629   __ br(Assembler::always, false, Assembler::pt, resolved);
  3630   __ delayed()->nop();
  3633   // Extract target class from constant pool
  3634   __ bind(quicked);
  3635   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3636   __ get_constant_pool(Lscratch);
  3637   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3638   __ bind(resolved);
  3639   __ load_klass(Otos_i, RobjKlass); // get value klass
  3641   // Generate a fast subtype check.  Branch to cast_ok if no
  3642   // failure.  Return 0 if failure.
  3643   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
  3644   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
  3645   // Not a subtype; return 0;
  3646   __ clr( Otos_i );
  3648   if (ProfileInterpreter) {
  3649     __ ba(false, done);
  3650     __ delayed()->nop();
  3652   __ bind(is_null);
  3653   __ profile_null_seen(G3_scratch);
  3654   __ bind(done);
  3657 void TemplateTable::_breakpoint() {
  3659    // Note: We get here even if we are single stepping..
  3660    // jbug inists on setting breakpoints at every bytecode
  3661    // even if we are in single step mode.
  3663    transition(vtos, vtos);
  3664    // get the unpatched byte code
  3665    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
  3666    __ mov(O0, Lbyte_code);
  3668    // post the breakpoint event
  3669    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
  3671    // complete the execution of original bytecode
  3672    __ dispatch_normal(vtos);
  3676 //----------------------------------------------------------------------------------------------------
  3677 // Exceptions
  3679 void TemplateTable::athrow() {
  3680   transition(atos, vtos);
  3682   // This works because exception is cached in Otos_i which is same as O0,
  3683   // which is same as what throw_exception_entry_expects
  3684   assert(Otos_i == Oexception, "see explanation above");
  3686   __ verify_oop(Otos_i);
  3687   __ null_check(Otos_i);
  3688   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
  3692 //----------------------------------------------------------------------------------------------------
  3693 // Synchronization
  3696 // See frame_sparc.hpp for monitor block layout.
  3697 // Monitor elements are dynamically allocated by growing stack as needed.
  3699 void TemplateTable::monitorenter() {
  3700   transition(atos, vtos);
  3701   __ verify_oop(Otos_i);
  3702   // Try to acquire a lock on the object
  3703   // Repeat until succeeded (i.e., until
  3704   // monitorenter returns true).
  3706   {   Label ok;
  3707     __ tst(Otos_i);
  3708     __ throw_if_not_1_x( Assembler::notZero,  ok);
  3709     __ delayed()->mov(Otos_i, Lscratch); // save obj
  3710     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3713   assert(O0 == Otos_i, "Be sure where the object to lock is");
  3715   // find a free slot in the monitor block
  3718   // initialize entry pointer
  3719   __ clr(O1); // points to free slot or NULL
  3722     Label entry, loop, exit;
  3723     __ add( __ top_most_monitor(), O2 ); // last one to check
  3724     __ ba( false, entry );
  3725     __ delayed()->mov( Lmonitors, O3 ); // first one to check
  3728     __ bind( loop );
  3730     __ verify_oop(O4);          // verify each monitor's oop
  3731     __ tst(O4); // is this entry unused?
  3732     if (VM_Version::v9_instructions_work())
  3733       __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
  3734     else {
  3735       Label L;
  3736       __ br( Assembler::zero, true, Assembler::pn, L );
  3737       __ delayed()->mov(O3, O1); // rememeber this one if match
  3738       __ bind(L);
  3741     __ cmp(O4, O0); // check if current entry is for same object
  3742     __ brx( Assembler::equal, false, Assembler::pn, exit );
  3743     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
  3745     __ bind( entry );
  3747     __ cmp( O3, O2 );
  3748     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3749     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
  3751     __ bind( exit );
  3754   { Label allocated;
  3756     // found free slot?
  3757     __ br_notnull(O1, false, Assembler::pn, allocated);
  3758     __ delayed()->nop();
  3760     __ add_monitor_to_stack( false, O2, O3 );
  3761     __ mov(Lmonitors, O1);
  3763     __ bind(allocated);
  3766   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
  3767   // The object has already been poped from the stack, so the expression stack looks correct.
  3768   __ inc(Lbcp);
  3770   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
  3771   __ lock_object(O1, O0);
  3773   // check if there's enough space on the stack for the monitors after locking
  3774   __ generate_stack_overflow_check(0);
  3776   // The bcp has already been incremented. Just need to dispatch to next instruction.
  3777   __ dispatch_next(vtos);
  3781 void TemplateTable::monitorexit() {
  3782   transition(atos, vtos);
  3783   __ verify_oop(Otos_i);
  3784   __ tst(Otos_i);
  3785   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
  3787   assert(O0 == Otos_i, "just checking");
  3789   { Label entry, loop, found;
  3790     __ add( __ top_most_monitor(), O2 ); // last one to check
  3791     __ ba(false, entry );
  3792     // use Lscratch to hold monitor elem to check, start with most recent monitor,
  3793     // By using a local it survives the call to the C routine.
  3794     __ delayed()->mov( Lmonitors, Lscratch );
  3796     __ bind( loop );
  3798     __ verify_oop(O4);          // verify each monitor's oop
  3799     __ cmp(O4, O0); // check if current entry is for desired object
  3800     __ brx( Assembler::equal, true, Assembler::pt, found );
  3801     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
  3803     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
  3805     __ bind( entry );
  3807     __ cmp( Lscratch, O2 );
  3808     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3809     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
  3811     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
  3812     __ should_not_reach_here();
  3814     __ bind(found);
  3816   __ unlock_object(O1);
  3820 //----------------------------------------------------------------------------------------------------
  3821 // Wide instructions
  3823 void TemplateTable::wide() {
  3824   transition(vtos, vtos);
  3825   __ ldub(Lbcp, 1, G3_scratch);// get next bc
  3826   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
  3827   AddressLiteral ep(Interpreter::_wentry_point);
  3828   __ set(ep, G4_scratch);
  3829   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
  3830   __ jmp(G3_scratch, G0);
  3831   __ delayed()->nop();
  3832   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
  3836 //----------------------------------------------------------------------------------------------------
  3837 // Multi arrays
  3839 void TemplateTable::multianewarray() {
  3840   transition(vtos, atos);
  3841      // put ndims * wordSize into Lscratch
  3842   __ ldub( Lbcp,     3,               Lscratch);
  3843   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
  3844      // Lesp points past last_dim, so set to O1 to first_dim address
  3845   __ add(  Lesp,     Lscratch,        O1);
  3846      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
  3847   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
  3849 #endif /* !CC_INTERP */

mercurial